1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Routines having to do with the 'struct sk_buff' memory handlers.
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Florian La Roche <rzsfl@rz.uni-sb.de>
9 * Alan Cox : Fixed the worst of the load
11 * Dave Platt : Interrupt stacking fix.
12 * Richard Kooijman : Timestamp fixes.
13 * Alan Cox : Changed buffer format.
14 * Alan Cox : destructor hook for AF_UNIX etc.
15 * Linus Torvalds : Better skb_clone.
16 * Alan Cox : Added skb_copy.
17 * Alan Cox : Added all the changed routines Linus
18 * only put in the headers
19 * Ray VanTassle : Fixed --skb->lock in free
20 * Alan Cox : skb_copy copy arp field
21 * Andi Kleen : slabified it.
22 * Robert Olsson : Removed skb_head_pool
25 * The __skb_ routines should be called with interrupts
26 * disabled, or you better be *real* sure that the operation is atomic
27 * with respect to whatever list is being frobbed (e.g. via lock_sock()
28 * or via disabling bottom half handlers, etc).
32 * The functions in this file will not compile correctly with gcc 2.4.x
35 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/kernel.h>
41 #include <linux/interrupt.h>
43 #include <linux/inet.h>
44 #include <linux/slab.h>
45 #include <linux/tcp.h>
46 #include <linux/udp.h>
47 #include <linux/sctp.h>
48 #include <linux/netdevice.h>
49 #ifdef CONFIG_NET_CLS_ACT
50 #include <net/pkt_sched.h>
52 #include <linux/string.h>
53 #include <linux/skbuff.h>
54 #include <linux/splice.h>
55 #include <linux/cache.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/init.h>
58 #include <linux/scatterlist.h>
59 #include <linux/errqueue.h>
60 #include <linux/prefetch.h>
61 #include <linux/if_vlan.h>
62 #include <linux/mpls.h>
63 #include <linux/kcov.h>
65 #include <net/protocol.h>
68 #include <net/checksum.h>
69 #include <net/ip6_checksum.h>
72 #include <net/mptcp.h>
73 #include <net/page_pool.h>
75 #include <linux/uaccess.h>
76 #include <trace/events/skb.h>
77 #include <linux/highmem.h>
78 #include <linux/capability.h>
79 #include <linux/user_namespace.h>
80 #include <linux/indirect_call_wrapper.h>
84 struct kmem_cache *skbuff_head_cache __ro_after_init;
85 static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
86 #ifdef CONFIG_SKB_EXTENSIONS
87 static struct kmem_cache *skbuff_ext_cache __ro_after_init;
89 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
90 EXPORT_SYMBOL(sysctl_max_skb_frags);
93 * skb_panic - private function for out-of-line support
97 * @msg: skb_over_panic or skb_under_panic
99 * Out-of-line support for skb_put() and skb_push().
100 * Called via the wrapper skb_over_panic() or skb_under_panic().
101 * Keep out of line to prevent kernel bloat.
102 * __builtin_return_address is not used because it is not always reliable.
104 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
107 pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n",
108 msg, addr, skb->len, sz, skb->head, skb->data,
109 (unsigned long)skb->tail, (unsigned long)skb->end,
110 skb->dev ? skb->dev->name : "<NULL>");
114 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
116 skb_panic(skb, sz, addr, __func__);
119 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
121 skb_panic(skb, sz, addr, __func__);
124 #define NAPI_SKB_CACHE_SIZE 64
125 #define NAPI_SKB_CACHE_BULK 16
126 #define NAPI_SKB_CACHE_HALF (NAPI_SKB_CACHE_SIZE / 2)
128 struct napi_alloc_cache {
129 struct page_frag_cache page;
130 unsigned int skb_count;
131 void *skb_cache[NAPI_SKB_CACHE_SIZE];
134 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
135 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
137 static void *__alloc_frag_align(unsigned int fragsz, gfp_t gfp_mask,
138 unsigned int align_mask)
140 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
142 return page_frag_alloc_align(&nc->page, fragsz, gfp_mask, align_mask);
145 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
147 fragsz = SKB_DATA_ALIGN(fragsz);
149 return __alloc_frag_align(fragsz, GFP_ATOMIC, align_mask);
151 EXPORT_SYMBOL(__napi_alloc_frag_align);
153 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
155 struct page_frag_cache *nc;
158 fragsz = SKB_DATA_ALIGN(fragsz);
159 if (in_irq() || irqs_disabled()) {
160 nc = this_cpu_ptr(&netdev_alloc_cache);
161 data = page_frag_alloc_align(nc, fragsz, GFP_ATOMIC, align_mask);
164 data = __alloc_frag_align(fragsz, GFP_ATOMIC, align_mask);
169 EXPORT_SYMBOL(__netdev_alloc_frag_align);
171 static struct sk_buff *napi_skb_cache_get(void)
173 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
176 if (unlikely(!nc->skb_count))
177 nc->skb_count = kmem_cache_alloc_bulk(skbuff_head_cache,
181 if (unlikely(!nc->skb_count))
184 skb = nc->skb_cache[--nc->skb_count];
185 kasan_unpoison_object_data(skbuff_head_cache, skb);
190 /* Caller must provide SKB that is memset cleared */
191 static void __build_skb_around(struct sk_buff *skb, void *data,
192 unsigned int frag_size)
194 struct skb_shared_info *shinfo;
195 unsigned int size = frag_size ? : ksize(data);
197 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
199 /* Assumes caller memset cleared SKB */
200 skb->truesize = SKB_TRUESIZE(size);
201 refcount_set(&skb->users, 1);
204 skb_reset_tail_pointer(skb);
205 skb->end = skb->tail + size;
206 skb->mac_header = (typeof(skb->mac_header))~0U;
207 skb->transport_header = (typeof(skb->transport_header))~0U;
209 /* make sure we initialize shinfo sequentially */
210 shinfo = skb_shinfo(skb);
211 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
212 atomic_set(&shinfo->dataref, 1);
214 skb_set_kcov_handle(skb, kcov_common_handle());
218 * __build_skb - build a network buffer
219 * @data: data buffer provided by caller
220 * @frag_size: size of data, or 0 if head was kmalloced
222 * Allocate a new &sk_buff. Caller provides space holding head and
223 * skb_shared_info. @data must have been allocated by kmalloc() only if
224 * @frag_size is 0, otherwise data should come from the page allocator
226 * The return is the new skb buffer.
227 * On a failure the return is %NULL, and @data is not freed.
229 * Before IO, driver allocates only data buffer where NIC put incoming frame
230 * Driver should add room at head (NET_SKB_PAD) and
231 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
232 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
233 * before giving packet to stack.
234 * RX rings only contains data buffers, not full skbs.
236 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
240 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
244 memset(skb, 0, offsetof(struct sk_buff, tail));
245 __build_skb_around(skb, data, frag_size);
250 /* build_skb() is wrapper over __build_skb(), that specifically
251 * takes care of skb->head and skb->pfmemalloc
252 * This means that if @frag_size is not zero, then @data must be backed
253 * by a page fragment, not kmalloc() or vmalloc()
255 struct sk_buff *build_skb(void *data, unsigned int frag_size)
257 struct sk_buff *skb = __build_skb(data, frag_size);
259 if (skb && frag_size) {
261 if (page_is_pfmemalloc(virt_to_head_page(data)))
266 EXPORT_SYMBOL(build_skb);
269 * build_skb_around - build a network buffer around provided skb
270 * @skb: sk_buff provide by caller, must be memset cleared
271 * @data: data buffer provided by caller
272 * @frag_size: size of data, or 0 if head was kmalloced
274 struct sk_buff *build_skb_around(struct sk_buff *skb,
275 void *data, unsigned int frag_size)
280 __build_skb_around(skb, data, frag_size);
284 if (page_is_pfmemalloc(virt_to_head_page(data)))
289 EXPORT_SYMBOL(build_skb_around);
292 * __napi_build_skb - build a network buffer
293 * @data: data buffer provided by caller
294 * @frag_size: size of data, or 0 if head was kmalloced
296 * Version of __build_skb() that uses NAPI percpu caches to obtain
297 * skbuff_head instead of inplace allocation.
299 * Returns a new &sk_buff on success, %NULL on allocation failure.
301 static struct sk_buff *__napi_build_skb(void *data, unsigned int frag_size)
305 skb = napi_skb_cache_get();
309 memset(skb, 0, offsetof(struct sk_buff, tail));
310 __build_skb_around(skb, data, frag_size);
316 * napi_build_skb - build a network buffer
317 * @data: data buffer provided by caller
318 * @frag_size: size of data, or 0 if head was kmalloced
320 * Version of __napi_build_skb() that takes care of skb->head_frag
321 * and skb->pfmemalloc when the data is a page or page fragment.
323 * Returns a new &sk_buff on success, %NULL on allocation failure.
325 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size)
327 struct sk_buff *skb = __napi_build_skb(data, frag_size);
329 if (likely(skb) && frag_size) {
331 skb_propagate_pfmemalloc(virt_to_head_page(data), skb);
336 EXPORT_SYMBOL(napi_build_skb);
339 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
340 * the caller if emergency pfmemalloc reserves are being used. If it is and
341 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
342 * may be used. Otherwise, the packet data may be discarded until enough
345 static void *kmalloc_reserve(size_t size, gfp_t flags, int node,
349 bool ret_pfmemalloc = false;
352 * Try a regular allocation, when that fails and we're not entitled
353 * to the reserves, fail.
355 obj = kmalloc_node_track_caller(size,
356 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
358 if (obj || !(gfp_pfmemalloc_allowed(flags)))
361 /* Try again but now we are using pfmemalloc reserves */
362 ret_pfmemalloc = true;
363 obj = kmalloc_node_track_caller(size, flags, node);
367 *pfmemalloc = ret_pfmemalloc;
372 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
373 * 'private' fields and also do memory statistics to find all the
379 * __alloc_skb - allocate a network buffer
380 * @size: size to allocate
381 * @gfp_mask: allocation mask
382 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
383 * instead of head cache and allocate a cloned (child) skb.
384 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
385 * allocations in case the data is required for writeback
386 * @node: numa node to allocate memory on
388 * Allocate a new &sk_buff. The returned buffer has no headroom and a
389 * tail room of at least size bytes. The object has a reference count
390 * of one. The return is the buffer. On a failure the return is %NULL.
392 * Buffers may only be allocated from interrupts using a @gfp_mask of
395 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
398 struct kmem_cache *cache;
403 cache = (flags & SKB_ALLOC_FCLONE)
404 ? skbuff_fclone_cache : skbuff_head_cache;
406 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
407 gfp_mask |= __GFP_MEMALLOC;
410 if ((flags & (SKB_ALLOC_FCLONE | SKB_ALLOC_NAPI)) == SKB_ALLOC_NAPI &&
411 likely(node == NUMA_NO_NODE || node == numa_mem_id()))
412 skb = napi_skb_cache_get();
414 skb = kmem_cache_alloc_node(cache, gfp_mask & ~GFP_DMA, node);
419 /* We do our best to align skb_shared_info on a separate cache
420 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
421 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
422 * Both skb->head and skb_shared_info are cache line aligned.
424 size = SKB_DATA_ALIGN(size);
425 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
426 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
429 /* kmalloc(size) might give us more room than requested.
430 * Put skb_shared_info exactly at the end of allocated zone,
431 * to allow max possible filling before reallocation.
433 size = SKB_WITH_OVERHEAD(ksize(data));
434 prefetchw(data + size);
437 * Only clear those fields we need to clear, not those that we will
438 * actually initialise below. Hence, don't put any more fields after
439 * the tail pointer in struct sk_buff!
441 memset(skb, 0, offsetof(struct sk_buff, tail));
442 __build_skb_around(skb, data, 0);
443 skb->pfmemalloc = pfmemalloc;
445 if (flags & SKB_ALLOC_FCLONE) {
446 struct sk_buff_fclones *fclones;
448 fclones = container_of(skb, struct sk_buff_fclones, skb1);
450 skb->fclone = SKB_FCLONE_ORIG;
451 refcount_set(&fclones->fclone_ref, 1);
453 fclones->skb2.fclone = SKB_FCLONE_CLONE;
459 kmem_cache_free(cache, skb);
462 EXPORT_SYMBOL(__alloc_skb);
465 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
466 * @dev: network device to receive on
467 * @len: length to allocate
468 * @gfp_mask: get_free_pages mask, passed to alloc_skb
470 * Allocate a new &sk_buff and assign it a usage count of one. The
471 * buffer has NET_SKB_PAD headroom built in. Users should allocate
472 * the headroom they think they need without accounting for the
473 * built in space. The built in space is used for optimisations.
475 * %NULL is returned if there is no free memory.
477 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
480 struct page_frag_cache *nc;
487 /* If requested length is either too small or too big,
488 * we use kmalloc() for skb->head allocation.
490 if (len <= SKB_WITH_OVERHEAD(1024) ||
491 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
492 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
493 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
499 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
500 len = SKB_DATA_ALIGN(len);
502 if (sk_memalloc_socks())
503 gfp_mask |= __GFP_MEMALLOC;
505 if (in_irq() || irqs_disabled()) {
506 nc = this_cpu_ptr(&netdev_alloc_cache);
507 data = page_frag_alloc(nc, len, gfp_mask);
508 pfmemalloc = nc->pfmemalloc;
511 nc = this_cpu_ptr(&napi_alloc_cache.page);
512 data = page_frag_alloc(nc, len, gfp_mask);
513 pfmemalloc = nc->pfmemalloc;
520 skb = __build_skb(data, len);
521 if (unlikely(!skb)) {
531 skb_reserve(skb, NET_SKB_PAD);
537 EXPORT_SYMBOL(__netdev_alloc_skb);
540 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
541 * @napi: napi instance this buffer was allocated for
542 * @len: length to allocate
543 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
545 * Allocate a new sk_buff for use in NAPI receive. This buffer will
546 * attempt to allocate the head from a special reserved region used
547 * only for NAPI Rx allocation. By doing this we can save several
548 * CPU cycles by avoiding having to disable and re-enable IRQs.
550 * %NULL is returned if there is no free memory.
552 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
555 struct napi_alloc_cache *nc;
559 len += NET_SKB_PAD + NET_IP_ALIGN;
561 /* If requested length is either too small or too big,
562 * we use kmalloc() for skb->head allocation.
564 if (len <= SKB_WITH_OVERHEAD(1024) ||
565 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
566 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
567 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI,
574 nc = this_cpu_ptr(&napi_alloc_cache);
575 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
576 len = SKB_DATA_ALIGN(len);
578 if (sk_memalloc_socks())
579 gfp_mask |= __GFP_MEMALLOC;
581 data = page_frag_alloc(&nc->page, len, gfp_mask);
585 skb = __napi_build_skb(data, len);
586 if (unlikely(!skb)) {
591 if (nc->page.pfmemalloc)
596 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
597 skb->dev = napi->dev;
602 EXPORT_SYMBOL(__napi_alloc_skb);
604 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
605 int size, unsigned int truesize)
607 skb_fill_page_desc(skb, i, page, off, size);
609 skb->data_len += size;
610 skb->truesize += truesize;
612 EXPORT_SYMBOL(skb_add_rx_frag);
614 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
615 unsigned int truesize)
617 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
619 skb_frag_size_add(frag, size);
621 skb->data_len += size;
622 skb->truesize += truesize;
624 EXPORT_SYMBOL(skb_coalesce_rx_frag);
626 static void skb_drop_list(struct sk_buff **listp)
628 kfree_skb_list(*listp);
632 static inline void skb_drop_fraglist(struct sk_buff *skb)
634 skb_drop_list(&skb_shinfo(skb)->frag_list);
637 static void skb_clone_fraglist(struct sk_buff *skb)
639 struct sk_buff *list;
641 skb_walk_frags(skb, list)
645 static void skb_free_head(struct sk_buff *skb)
647 unsigned char *head = skb->head;
649 if (skb->head_frag) {
650 if (skb_pp_recycle(skb, head))
658 static void skb_release_data(struct sk_buff *skb)
660 struct skb_shared_info *shinfo = skb_shinfo(skb);
664 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
668 skb_zcopy_clear(skb, true);
670 for (i = 0; i < shinfo->nr_frags; i++)
671 __skb_frag_unref(&shinfo->frags[i], skb->pp_recycle);
673 if (shinfo->frag_list)
674 kfree_skb_list(shinfo->frag_list);
678 /* When we clone an SKB we copy the reycling bit. The pp_recycle
679 * bit is only set on the head though, so in order to avoid races
680 * while trying to recycle fragments on __skb_frag_unref() we need
681 * to make one SKB responsible for triggering the recycle path.
682 * So disable the recycling bit if an SKB is cloned and we have
683 * additional references to to the fragmented part of the SKB.
684 * Eventually the last SKB will have the recycling bit set and it's
685 * dataref set to 0, which will trigger the recycling
691 * Free an skbuff by memory without cleaning the state.
693 static void kfree_skbmem(struct sk_buff *skb)
695 struct sk_buff_fclones *fclones;
697 switch (skb->fclone) {
698 case SKB_FCLONE_UNAVAILABLE:
699 kmem_cache_free(skbuff_head_cache, skb);
702 case SKB_FCLONE_ORIG:
703 fclones = container_of(skb, struct sk_buff_fclones, skb1);
705 /* We usually free the clone (TX completion) before original skb
706 * This test would have no chance to be true for the clone,
707 * while here, branch prediction will be good.
709 if (refcount_read(&fclones->fclone_ref) == 1)
713 default: /* SKB_FCLONE_CLONE */
714 fclones = container_of(skb, struct sk_buff_fclones, skb2);
717 if (!refcount_dec_and_test(&fclones->fclone_ref))
720 kmem_cache_free(skbuff_fclone_cache, fclones);
723 void skb_release_head_state(struct sk_buff *skb)
726 if (skb->destructor) {
728 skb->destructor(skb);
730 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
731 nf_conntrack_put(skb_nfct(skb));
736 /* Free everything but the sk_buff shell. */
737 static void skb_release_all(struct sk_buff *skb)
739 skb_release_head_state(skb);
740 if (likely(skb->head))
741 skb_release_data(skb);
745 * __kfree_skb - private function
748 * Free an sk_buff. Release anything attached to the buffer.
749 * Clean the state. This is an internal helper function. Users should
750 * always call kfree_skb
753 void __kfree_skb(struct sk_buff *skb)
755 skb_release_all(skb);
758 EXPORT_SYMBOL(__kfree_skb);
761 * kfree_skb - free an sk_buff
762 * @skb: buffer to free
764 * Drop a reference to the buffer and free it if the usage count has
767 void kfree_skb(struct sk_buff *skb)
772 trace_kfree_skb(skb, __builtin_return_address(0));
775 EXPORT_SYMBOL(kfree_skb);
777 void kfree_skb_list(struct sk_buff *segs)
780 struct sk_buff *next = segs->next;
786 EXPORT_SYMBOL(kfree_skb_list);
788 /* Dump skb information and contents.
790 * Must only be called from net_ratelimit()-ed paths.
792 * Dumps whole packets if full_pkt, only headers otherwise.
794 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
796 struct skb_shared_info *sh = skb_shinfo(skb);
797 struct net_device *dev = skb->dev;
798 struct sock *sk = skb->sk;
799 struct sk_buff *list_skb;
800 bool has_mac, has_trans;
801 int headroom, tailroom;
807 len = min_t(int, skb->len, MAX_HEADER + 128);
809 headroom = skb_headroom(skb);
810 tailroom = skb_tailroom(skb);
812 has_mac = skb_mac_header_was_set(skb);
813 has_trans = skb_transport_header_was_set(skb);
815 printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
816 "mac=(%d,%d) net=(%d,%d) trans=%d\n"
817 "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
818 "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
819 "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
820 level, skb->len, headroom, skb_headlen(skb), tailroom,
821 has_mac ? skb->mac_header : -1,
822 has_mac ? skb_mac_header_len(skb) : -1,
824 has_trans ? skb_network_header_len(skb) : -1,
825 has_trans ? skb->transport_header : -1,
826 sh->tx_flags, sh->nr_frags,
827 sh->gso_size, sh->gso_type, sh->gso_segs,
828 skb->csum, skb->ip_summed, skb->csum_complete_sw,
829 skb->csum_valid, skb->csum_level,
830 skb->hash, skb->sw_hash, skb->l4_hash,
831 ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
834 printk("%sdev name=%s feat=0x%pNF\n",
835 level, dev->name, &dev->features);
837 printk("%ssk family=%hu type=%u proto=%u\n",
838 level, sk->sk_family, sk->sk_type, sk->sk_protocol);
840 if (full_pkt && headroom)
841 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
842 16, 1, skb->head, headroom, false);
844 seg_len = min_t(int, skb_headlen(skb), len);
846 print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET,
847 16, 1, skb->data, seg_len, false);
850 if (full_pkt && tailroom)
851 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
852 16, 1, skb_tail_pointer(skb), tailroom, false);
854 for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
855 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
856 u32 p_off, p_len, copied;
860 skb_frag_foreach_page(frag, skb_frag_off(frag),
861 skb_frag_size(frag), p, p_off, p_len,
863 seg_len = min_t(int, p_len, len);
864 vaddr = kmap_atomic(p);
865 print_hex_dump(level, "skb frag: ",
867 16, 1, vaddr + p_off, seg_len, false);
868 kunmap_atomic(vaddr);
875 if (full_pkt && skb_has_frag_list(skb)) {
876 printk("skb fraglist:\n");
877 skb_walk_frags(skb, list_skb)
878 skb_dump(level, list_skb, true);
881 EXPORT_SYMBOL(skb_dump);
884 * skb_tx_error - report an sk_buff xmit error
885 * @skb: buffer that triggered an error
887 * Report xmit error if a device callback is tracking this skb.
888 * skb must be freed afterwards.
890 void skb_tx_error(struct sk_buff *skb)
892 skb_zcopy_clear(skb, true);
894 EXPORT_SYMBOL(skb_tx_error);
896 #ifdef CONFIG_TRACEPOINTS
898 * consume_skb - free an skbuff
899 * @skb: buffer to free
901 * Drop a ref to the buffer and free it if the usage count has hit zero
902 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
903 * is being dropped after a failure and notes that
905 void consume_skb(struct sk_buff *skb)
910 trace_consume_skb(skb);
913 EXPORT_SYMBOL(consume_skb);
917 * __consume_stateless_skb - free an skbuff, assuming it is stateless
918 * @skb: buffer to free
920 * Alike consume_skb(), but this variant assumes that this is the last
921 * skb reference and all the head states have been already dropped
923 void __consume_stateless_skb(struct sk_buff *skb)
925 trace_consume_skb(skb);
926 skb_release_data(skb);
930 static void napi_skb_cache_put(struct sk_buff *skb)
932 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
935 kasan_poison_object_data(skbuff_head_cache, skb);
936 nc->skb_cache[nc->skb_count++] = skb;
938 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
939 for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++)
940 kasan_unpoison_object_data(skbuff_head_cache,
943 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_HALF,
944 nc->skb_cache + NAPI_SKB_CACHE_HALF);
945 nc->skb_count = NAPI_SKB_CACHE_HALF;
949 void __kfree_skb_defer(struct sk_buff *skb)
951 skb_release_all(skb);
952 napi_skb_cache_put(skb);
955 void napi_skb_free_stolen_head(struct sk_buff *skb)
960 napi_skb_cache_put(skb);
963 void napi_consume_skb(struct sk_buff *skb, int budget)
965 /* Zero budget indicate non-NAPI context called us, like netpoll */
966 if (unlikely(!budget)) {
967 dev_consume_skb_any(skb);
971 lockdep_assert_in_softirq();
976 /* if reaching here SKB is ready to free */
977 trace_consume_skb(skb);
979 /* if SKB is a clone, don't handle this case */
980 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
985 skb_release_all(skb);
986 napi_skb_cache_put(skb);
988 EXPORT_SYMBOL(napi_consume_skb);
990 /* Make sure a field is enclosed inside headers_start/headers_end section */
991 #define CHECK_SKB_FIELD(field) \
992 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
993 offsetof(struct sk_buff, headers_start)); \
994 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
995 offsetof(struct sk_buff, headers_end)); \
997 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
999 new->tstamp = old->tstamp;
1000 /* We do not copy old->sk */
1001 new->dev = old->dev;
1002 memcpy(new->cb, old->cb, sizeof(old->cb));
1003 skb_dst_copy(new, old);
1004 __skb_ext_copy(new, old);
1005 __nf_copy(new, old, false);
1007 /* Note : this field could be in headers_start/headers_end section
1008 * It is not yet because we do not want to have a 16 bit hole
1010 new->queue_mapping = old->queue_mapping;
1012 memcpy(&new->headers_start, &old->headers_start,
1013 offsetof(struct sk_buff, headers_end) -
1014 offsetof(struct sk_buff, headers_start));
1015 CHECK_SKB_FIELD(protocol);
1016 CHECK_SKB_FIELD(csum);
1017 CHECK_SKB_FIELD(hash);
1018 CHECK_SKB_FIELD(priority);
1019 CHECK_SKB_FIELD(skb_iif);
1020 CHECK_SKB_FIELD(vlan_proto);
1021 CHECK_SKB_FIELD(vlan_tci);
1022 CHECK_SKB_FIELD(transport_header);
1023 CHECK_SKB_FIELD(network_header);
1024 CHECK_SKB_FIELD(mac_header);
1025 CHECK_SKB_FIELD(inner_protocol);
1026 CHECK_SKB_FIELD(inner_transport_header);
1027 CHECK_SKB_FIELD(inner_network_header);
1028 CHECK_SKB_FIELD(inner_mac_header);
1029 CHECK_SKB_FIELD(mark);
1030 #ifdef CONFIG_NETWORK_SECMARK
1031 CHECK_SKB_FIELD(secmark);
1033 #ifdef CONFIG_NET_RX_BUSY_POLL
1034 CHECK_SKB_FIELD(napi_id);
1037 CHECK_SKB_FIELD(sender_cpu);
1039 #ifdef CONFIG_NET_SCHED
1040 CHECK_SKB_FIELD(tc_index);
1046 * You should not add any new code to this function. Add it to
1047 * __copy_skb_header above instead.
1049 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
1051 #define C(x) n->x = skb->x
1053 n->next = n->prev = NULL;
1055 __copy_skb_header(n, skb);
1060 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
1066 n->destructor = NULL;
1073 refcount_set(&n->users, 1);
1075 atomic_inc(&(skb_shinfo(skb)->dataref));
1083 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1084 * @first: first sk_buff of the msg
1086 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1090 n = alloc_skb(0, GFP_ATOMIC);
1094 n->len = first->len;
1095 n->data_len = first->len;
1096 n->truesize = first->truesize;
1098 skb_shinfo(n)->frag_list = first;
1100 __copy_skb_header(n, first);
1101 n->destructor = NULL;
1105 EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1108 * skb_morph - morph one skb into another
1109 * @dst: the skb to receive the contents
1110 * @src: the skb to supply the contents
1112 * This is identical to skb_clone except that the target skb is
1113 * supplied by the user.
1115 * The target skb is returned upon exit.
1117 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1119 skb_release_all(dst);
1120 return __skb_clone(dst, src);
1122 EXPORT_SYMBOL_GPL(skb_morph);
1124 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1126 unsigned long max_pg, num_pg, new_pg, old_pg;
1127 struct user_struct *user;
1129 if (capable(CAP_IPC_LOCK) || !size)
1132 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
1133 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1134 user = mmp->user ? : current_user();
1137 old_pg = atomic_long_read(&user->locked_vm);
1138 new_pg = old_pg + num_pg;
1139 if (new_pg > max_pg)
1141 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
1145 mmp->user = get_uid(user);
1146 mmp->num_pg = num_pg;
1148 mmp->num_pg += num_pg;
1153 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1155 void mm_unaccount_pinned_pages(struct mmpin *mmp)
1158 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1159 free_uid(mmp->user);
1162 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1164 struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size)
1166 struct ubuf_info *uarg;
1167 struct sk_buff *skb;
1169 WARN_ON_ONCE(!in_task());
1171 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1175 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1176 uarg = (void *)skb->cb;
1177 uarg->mmp.user = NULL;
1179 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1184 uarg->callback = msg_zerocopy_callback;
1185 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1187 uarg->bytelen = size;
1189 uarg->flags = SKBFL_ZEROCOPY_FRAG;
1190 refcount_set(&uarg->refcnt, 1);
1195 EXPORT_SYMBOL_GPL(msg_zerocopy_alloc);
1197 static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
1199 return container_of((void *)uarg, struct sk_buff, cb);
1202 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1203 struct ubuf_info *uarg)
1206 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
1209 /* realloc only when socket is locked (TCP, UDP cork),
1210 * so uarg->len and sk_zckey access is serialized
1212 if (!sock_owned_by_user(sk)) {
1217 bytelen = uarg->bytelen + size;
1218 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1219 /* TCP can create new skb to attach new uarg */
1220 if (sk->sk_type == SOCK_STREAM)
1225 next = (u32)atomic_read(&sk->sk_zckey);
1226 if ((u32)(uarg->id + uarg->len) == next) {
1227 if (mm_account_pinned_pages(&uarg->mmp, size))
1230 uarg->bytelen = bytelen;
1231 atomic_set(&sk->sk_zckey, ++next);
1233 /* no extra ref when appending to datagram (MSG_MORE) */
1234 if (sk->sk_type == SOCK_STREAM)
1235 net_zcopy_get(uarg);
1242 return msg_zerocopy_alloc(sk, size);
1244 EXPORT_SYMBOL_GPL(msg_zerocopy_realloc);
1246 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1248 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1252 old_lo = serr->ee.ee_info;
1253 old_hi = serr->ee.ee_data;
1254 sum_len = old_hi - old_lo + 1ULL + len;
1256 if (sum_len >= (1ULL << 32))
1259 if (lo != old_hi + 1)
1262 serr->ee.ee_data += len;
1266 static void __msg_zerocopy_callback(struct ubuf_info *uarg)
1268 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1269 struct sock_exterr_skb *serr;
1270 struct sock *sk = skb->sk;
1271 struct sk_buff_head *q;
1272 unsigned long flags;
1277 mm_unaccount_pinned_pages(&uarg->mmp);
1279 /* if !len, there was only 1 call, and it was aborted
1280 * so do not queue a completion notification
1282 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1287 hi = uarg->id + len - 1;
1288 is_zerocopy = uarg->zerocopy;
1290 serr = SKB_EXT_ERR(skb);
1291 memset(serr, 0, sizeof(*serr));
1292 serr->ee.ee_errno = 0;
1293 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1294 serr->ee.ee_data = hi;
1295 serr->ee.ee_info = lo;
1297 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1299 q = &sk->sk_error_queue;
1300 spin_lock_irqsave(&q->lock, flags);
1301 tail = skb_peek_tail(q);
1302 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1303 !skb_zerocopy_notify_extend(tail, lo, len)) {
1304 __skb_queue_tail(q, skb);
1307 spin_unlock_irqrestore(&q->lock, flags);
1309 sk_error_report(sk);
1316 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1319 uarg->zerocopy = uarg->zerocopy & success;
1321 if (refcount_dec_and_test(&uarg->refcnt))
1322 __msg_zerocopy_callback(uarg);
1324 EXPORT_SYMBOL_GPL(msg_zerocopy_callback);
1326 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1328 struct sock *sk = skb_from_uarg(uarg)->sk;
1330 atomic_dec(&sk->sk_zckey);
1334 msg_zerocopy_callback(NULL, uarg, true);
1336 EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort);
1338 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len)
1340 return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
1342 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram);
1344 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1345 struct msghdr *msg, int len,
1346 struct ubuf_info *uarg)
1348 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1349 struct iov_iter orig_iter = msg->msg_iter;
1350 int err, orig_len = skb->len;
1352 /* An skb can only point to one uarg. This edge case happens when
1353 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1355 if (orig_uarg && uarg != orig_uarg)
1358 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1359 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1360 struct sock *save_sk = skb->sk;
1362 /* Streams do not free skb on error. Reset to prev state. */
1363 msg->msg_iter = orig_iter;
1365 ___pskb_trim(skb, orig_len);
1370 skb_zcopy_set(skb, uarg, NULL);
1371 return skb->len - orig_len;
1373 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1375 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1378 if (skb_zcopy(orig)) {
1379 if (skb_zcopy(nskb)) {
1380 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1385 if (skb_uarg(nskb) == skb_uarg(orig))
1387 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1390 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1396 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1397 * @skb: the skb to modify
1398 * @gfp_mask: allocation priority
1400 * This must be called on skb with SKBFL_ZEROCOPY_ENABLE.
1401 * It will copy all frags into kernel and drop the reference
1402 * to userspace pages.
1404 * If this function is called from an interrupt gfp_mask() must be
1407 * Returns 0 on success or a negative error code on failure
1408 * to allocate kernel memory to copy to.
1410 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1412 int num_frags = skb_shinfo(skb)->nr_frags;
1413 struct page *page, *head = NULL;
1417 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1423 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1424 for (i = 0; i < new_frags; i++) {
1425 page = alloc_page(gfp_mask);
1428 struct page *next = (struct page *)page_private(head);
1434 set_page_private(page, (unsigned long)head);
1440 for (i = 0; i < num_frags; i++) {
1441 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1442 u32 p_off, p_len, copied;
1446 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1447 p, p_off, p_len, copied) {
1449 vaddr = kmap_atomic(p);
1451 while (done < p_len) {
1452 if (d_off == PAGE_SIZE) {
1454 page = (struct page *)page_private(page);
1456 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1457 memcpy(page_address(page) + d_off,
1458 vaddr + p_off + done, copy);
1462 kunmap_atomic(vaddr);
1466 /* skb frags release userspace buffers */
1467 for (i = 0; i < num_frags; i++)
1468 skb_frag_unref(skb, i);
1470 /* skb frags point to kernel buffers */
1471 for (i = 0; i < new_frags - 1; i++) {
1472 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1473 head = (struct page *)page_private(head);
1475 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1476 skb_shinfo(skb)->nr_frags = new_frags;
1479 skb_zcopy_clear(skb, false);
1482 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1485 * skb_clone - duplicate an sk_buff
1486 * @skb: buffer to clone
1487 * @gfp_mask: allocation priority
1489 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1490 * copies share the same packet data but not structure. The new
1491 * buffer has a reference count of 1. If the allocation fails the
1492 * function returns %NULL otherwise the new buffer is returned.
1494 * If this function is called from an interrupt gfp_mask() must be
1498 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1500 struct sk_buff_fclones *fclones = container_of(skb,
1501 struct sk_buff_fclones,
1505 if (skb_orphan_frags(skb, gfp_mask))
1508 if (skb->fclone == SKB_FCLONE_ORIG &&
1509 refcount_read(&fclones->fclone_ref) == 1) {
1511 refcount_set(&fclones->fclone_ref, 2);
1513 if (skb_pfmemalloc(skb))
1514 gfp_mask |= __GFP_MEMALLOC;
1516 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1520 n->fclone = SKB_FCLONE_UNAVAILABLE;
1523 return __skb_clone(n, skb);
1525 EXPORT_SYMBOL(skb_clone);
1527 void skb_headers_offset_update(struct sk_buff *skb, int off)
1529 /* Only adjust this if it actually is csum_start rather than csum */
1530 if (skb->ip_summed == CHECKSUM_PARTIAL)
1531 skb->csum_start += off;
1532 /* {transport,network,mac}_header and tail are relative to skb->head */
1533 skb->transport_header += off;
1534 skb->network_header += off;
1535 if (skb_mac_header_was_set(skb))
1536 skb->mac_header += off;
1537 skb->inner_transport_header += off;
1538 skb->inner_network_header += off;
1539 skb->inner_mac_header += off;
1541 EXPORT_SYMBOL(skb_headers_offset_update);
1543 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1545 __copy_skb_header(new, old);
1547 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1548 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1549 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1551 EXPORT_SYMBOL(skb_copy_header);
1553 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1555 if (skb_pfmemalloc(skb))
1556 return SKB_ALLOC_RX;
1561 * skb_copy - create private copy of an sk_buff
1562 * @skb: buffer to copy
1563 * @gfp_mask: allocation priority
1565 * Make a copy of both an &sk_buff and its data. This is used when the
1566 * caller wishes to modify the data and needs a private copy of the
1567 * data to alter. Returns %NULL on failure or the pointer to the buffer
1568 * on success. The returned buffer has a reference count of 1.
1570 * As by-product this function converts non-linear &sk_buff to linear
1571 * one, so that &sk_buff becomes completely private and caller is allowed
1572 * to modify all the data of returned buffer. This means that this
1573 * function is not recommended for use in circumstances when only
1574 * header is going to be modified. Use pskb_copy() instead.
1577 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1579 int headerlen = skb_headroom(skb);
1580 unsigned int size = skb_end_offset(skb) + skb->data_len;
1581 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1582 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1587 /* Set the data pointer */
1588 skb_reserve(n, headerlen);
1589 /* Set the tail pointer and length */
1590 skb_put(n, skb->len);
1592 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1594 skb_copy_header(n, skb);
1597 EXPORT_SYMBOL(skb_copy);
1600 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1601 * @skb: buffer to copy
1602 * @headroom: headroom of new skb
1603 * @gfp_mask: allocation priority
1604 * @fclone: if true allocate the copy of the skb from the fclone
1605 * cache instead of the head cache; it is recommended to set this
1606 * to true for the cases where the copy will likely be cloned
1608 * Make a copy of both an &sk_buff and part of its data, located
1609 * in header. Fragmented data remain shared. This is used when
1610 * the caller wishes to modify only header of &sk_buff and needs
1611 * private copy of the header to alter. Returns %NULL on failure
1612 * or the pointer to the buffer on success.
1613 * The returned buffer has a reference count of 1.
1616 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1617 gfp_t gfp_mask, bool fclone)
1619 unsigned int size = skb_headlen(skb) + headroom;
1620 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1621 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1626 /* Set the data pointer */
1627 skb_reserve(n, headroom);
1628 /* Set the tail pointer and length */
1629 skb_put(n, skb_headlen(skb));
1630 /* Copy the bytes */
1631 skb_copy_from_linear_data(skb, n->data, n->len);
1633 n->truesize += skb->data_len;
1634 n->data_len = skb->data_len;
1637 if (skb_shinfo(skb)->nr_frags) {
1640 if (skb_orphan_frags(skb, gfp_mask) ||
1641 skb_zerocopy_clone(n, skb, gfp_mask)) {
1646 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1647 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1648 skb_frag_ref(skb, i);
1650 skb_shinfo(n)->nr_frags = i;
1653 if (skb_has_frag_list(skb)) {
1654 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1655 skb_clone_fraglist(n);
1658 skb_copy_header(n, skb);
1662 EXPORT_SYMBOL(__pskb_copy_fclone);
1665 * pskb_expand_head - reallocate header of &sk_buff
1666 * @skb: buffer to reallocate
1667 * @nhead: room to add at head
1668 * @ntail: room to add at tail
1669 * @gfp_mask: allocation priority
1671 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1672 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1673 * reference count of 1. Returns zero in the case of success or error,
1674 * if expansion failed. In the last case, &sk_buff is not changed.
1676 * All the pointers pointing into skb header may change and must be
1677 * reloaded after call to this function.
1680 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1683 int i, osize = skb_end_offset(skb);
1684 int size = osize + nhead + ntail;
1690 BUG_ON(skb_shared(skb));
1692 size = SKB_DATA_ALIGN(size);
1694 if (skb_pfmemalloc(skb))
1695 gfp_mask |= __GFP_MEMALLOC;
1696 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1697 gfp_mask, NUMA_NO_NODE, NULL);
1700 size = SKB_WITH_OVERHEAD(ksize(data));
1702 /* Copy only real data... and, alas, header. This should be
1703 * optimized for the cases when header is void.
1705 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1707 memcpy((struct skb_shared_info *)(data + size),
1709 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1712 * if shinfo is shared we must drop the old head gracefully, but if it
1713 * is not we can just drop the old head and let the existing refcount
1714 * be since all we did is relocate the values
1716 if (skb_cloned(skb)) {
1717 if (skb_orphan_frags(skb, gfp_mask))
1720 refcount_inc(&skb_uarg(skb)->refcnt);
1721 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1722 skb_frag_ref(skb, i);
1724 if (skb_has_frag_list(skb))
1725 skb_clone_fraglist(skb);
1727 skb_release_data(skb);
1731 off = (data + nhead) - skb->head;
1736 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1740 skb->end = skb->head + size;
1743 skb_headers_offset_update(skb, nhead);
1747 atomic_set(&skb_shinfo(skb)->dataref, 1);
1749 skb_metadata_clear(skb);
1751 /* It is not generally safe to change skb->truesize.
1752 * For the moment, we really care of rx path, or
1753 * when skb is orphaned (not attached to a socket).
1755 if (!skb->sk || skb->destructor == sock_edemux)
1756 skb->truesize += size - osize;
1765 EXPORT_SYMBOL(pskb_expand_head);
1767 /* Make private copy of skb with writable head and some headroom */
1769 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1771 struct sk_buff *skb2;
1772 int delta = headroom - skb_headroom(skb);
1775 skb2 = pskb_copy(skb, GFP_ATOMIC);
1777 skb2 = skb_clone(skb, GFP_ATOMIC);
1778 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1786 EXPORT_SYMBOL(skb_realloc_headroom);
1789 * skb_copy_expand - copy and expand sk_buff
1790 * @skb: buffer to copy
1791 * @newheadroom: new free bytes at head
1792 * @newtailroom: new free bytes at tail
1793 * @gfp_mask: allocation priority
1795 * Make a copy of both an &sk_buff and its data and while doing so
1796 * allocate additional space.
1798 * This is used when the caller wishes to modify the data and needs a
1799 * private copy of the data to alter as well as more space for new fields.
1800 * Returns %NULL on failure or the pointer to the buffer
1801 * on success. The returned buffer has a reference count of 1.
1803 * You must pass %GFP_ATOMIC as the allocation priority if this function
1804 * is called from an interrupt.
1806 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1807 int newheadroom, int newtailroom,
1811 * Allocate the copy buffer
1813 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1814 gfp_mask, skb_alloc_rx_flag(skb),
1816 int oldheadroom = skb_headroom(skb);
1817 int head_copy_len, head_copy_off;
1822 skb_reserve(n, newheadroom);
1824 /* Set the tail pointer and length */
1825 skb_put(n, skb->len);
1827 head_copy_len = oldheadroom;
1829 if (newheadroom <= head_copy_len)
1830 head_copy_len = newheadroom;
1832 head_copy_off = newheadroom - head_copy_len;
1834 /* Copy the linear header and data. */
1835 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1836 skb->len + head_copy_len));
1838 skb_copy_header(n, skb);
1840 skb_headers_offset_update(n, newheadroom - oldheadroom);
1844 EXPORT_SYMBOL(skb_copy_expand);
1847 * __skb_pad - zero pad the tail of an skb
1848 * @skb: buffer to pad
1849 * @pad: space to pad
1850 * @free_on_error: free buffer on error
1852 * Ensure that a buffer is followed by a padding area that is zero
1853 * filled. Used by network drivers which may DMA or transfer data
1854 * beyond the buffer end onto the wire.
1856 * May return error in out of memory cases. The skb is freed on error
1857 * if @free_on_error is true.
1860 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1865 /* If the skbuff is non linear tailroom is always zero.. */
1866 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1867 memset(skb->data+skb->len, 0, pad);
1871 ntail = skb->data_len + pad - (skb->end - skb->tail);
1872 if (likely(skb_cloned(skb) || ntail > 0)) {
1873 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1878 /* FIXME: The use of this function with non-linear skb's really needs
1881 err = skb_linearize(skb);
1885 memset(skb->data + skb->len, 0, pad);
1893 EXPORT_SYMBOL(__skb_pad);
1896 * pskb_put - add data to the tail of a potentially fragmented buffer
1897 * @skb: start of the buffer to use
1898 * @tail: tail fragment of the buffer to use
1899 * @len: amount of data to add
1901 * This function extends the used data area of the potentially
1902 * fragmented buffer. @tail must be the last fragment of @skb -- or
1903 * @skb itself. If this would exceed the total buffer size the kernel
1904 * will panic. A pointer to the first byte of the extra data is
1908 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1911 skb->data_len += len;
1914 return skb_put(tail, len);
1916 EXPORT_SYMBOL_GPL(pskb_put);
1919 * skb_put - add data to a buffer
1920 * @skb: buffer to use
1921 * @len: amount of data to add
1923 * This function extends the used data area of the buffer. If this would
1924 * exceed the total buffer size the kernel will panic. A pointer to the
1925 * first byte of the extra data is returned.
1927 void *skb_put(struct sk_buff *skb, unsigned int len)
1929 void *tmp = skb_tail_pointer(skb);
1930 SKB_LINEAR_ASSERT(skb);
1933 if (unlikely(skb->tail > skb->end))
1934 skb_over_panic(skb, len, __builtin_return_address(0));
1937 EXPORT_SYMBOL(skb_put);
1940 * skb_push - add data to the start of a buffer
1941 * @skb: buffer to use
1942 * @len: amount of data to add
1944 * This function extends the used data area of the buffer at the buffer
1945 * start. If this would exceed the total buffer headroom the kernel will
1946 * panic. A pointer to the first byte of the extra data is returned.
1948 void *skb_push(struct sk_buff *skb, unsigned int len)
1952 if (unlikely(skb->data < skb->head))
1953 skb_under_panic(skb, len, __builtin_return_address(0));
1956 EXPORT_SYMBOL(skb_push);
1959 * skb_pull - remove data from the start of a buffer
1960 * @skb: buffer to use
1961 * @len: amount of data to remove
1963 * This function removes data from the start of a buffer, returning
1964 * the memory to the headroom. A pointer to the next data in the buffer
1965 * is returned. Once the data has been pulled future pushes will overwrite
1968 void *skb_pull(struct sk_buff *skb, unsigned int len)
1970 return skb_pull_inline(skb, len);
1972 EXPORT_SYMBOL(skb_pull);
1975 * skb_trim - remove end from a buffer
1976 * @skb: buffer to alter
1979 * Cut the length of a buffer down by removing data from the tail. If
1980 * the buffer is already under the length specified it is not modified.
1981 * The skb must be linear.
1983 void skb_trim(struct sk_buff *skb, unsigned int len)
1986 __skb_trim(skb, len);
1988 EXPORT_SYMBOL(skb_trim);
1990 /* Trims skb to length len. It can change skb pointers.
1993 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1995 struct sk_buff **fragp;
1996 struct sk_buff *frag;
1997 int offset = skb_headlen(skb);
1998 int nfrags = skb_shinfo(skb)->nr_frags;
2002 if (skb_cloned(skb) &&
2003 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
2010 for (; i < nfrags; i++) {
2011 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2018 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
2021 skb_shinfo(skb)->nr_frags = i;
2023 for (; i < nfrags; i++)
2024 skb_frag_unref(skb, i);
2026 if (skb_has_frag_list(skb))
2027 skb_drop_fraglist(skb);
2031 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
2032 fragp = &frag->next) {
2033 int end = offset + frag->len;
2035 if (skb_shared(frag)) {
2036 struct sk_buff *nfrag;
2038 nfrag = skb_clone(frag, GFP_ATOMIC);
2039 if (unlikely(!nfrag))
2042 nfrag->next = frag->next;
2054 unlikely((err = pskb_trim(frag, len - offset))))
2058 skb_drop_list(&frag->next);
2063 if (len > skb_headlen(skb)) {
2064 skb->data_len -= skb->len - len;
2069 skb_set_tail_pointer(skb, len);
2072 if (!skb->sk || skb->destructor == sock_edemux)
2076 EXPORT_SYMBOL(___pskb_trim);
2078 /* Note : use pskb_trim_rcsum() instead of calling this directly
2080 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2082 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2083 int delta = skb->len - len;
2085 skb->csum = csum_block_sub(skb->csum,
2086 skb_checksum(skb, len, delta, 0),
2088 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
2089 int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len;
2090 int offset = skb_checksum_start_offset(skb) + skb->csum_offset;
2092 if (offset + sizeof(__sum16) > hdlen)
2095 return __pskb_trim(skb, len);
2097 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2100 * __pskb_pull_tail - advance tail of skb header
2101 * @skb: buffer to reallocate
2102 * @delta: number of bytes to advance tail
2104 * The function makes a sense only on a fragmented &sk_buff,
2105 * it expands header moving its tail forward and copying necessary
2106 * data from fragmented part.
2108 * &sk_buff MUST have reference count of 1.
2110 * Returns %NULL (and &sk_buff does not change) if pull failed
2111 * or value of new tail of skb in the case of success.
2113 * All the pointers pointing into skb header may change and must be
2114 * reloaded after call to this function.
2117 /* Moves tail of skb head forward, copying data from fragmented part,
2118 * when it is necessary.
2119 * 1. It may fail due to malloc failure.
2120 * 2. It may change skb pointers.
2122 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2124 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2126 /* If skb has not enough free space at tail, get new one
2127 * plus 128 bytes for future expansions. If we have enough
2128 * room at tail, reallocate without expansion only if skb is cloned.
2130 int i, k, eat = (skb->tail + delta) - skb->end;
2132 if (eat > 0 || skb_cloned(skb)) {
2133 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2138 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2139 skb_tail_pointer(skb), delta));
2141 /* Optimization: no fragments, no reasons to preestimate
2142 * size of pulled pages. Superb.
2144 if (!skb_has_frag_list(skb))
2147 /* Estimate size of pulled pages. */
2149 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2150 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2157 /* If we need update frag list, we are in troubles.
2158 * Certainly, it is possible to add an offset to skb data,
2159 * but taking into account that pulling is expected to
2160 * be very rare operation, it is worth to fight against
2161 * further bloating skb head and crucify ourselves here instead.
2162 * Pure masohism, indeed. 8)8)
2165 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2166 struct sk_buff *clone = NULL;
2167 struct sk_buff *insp = NULL;
2170 if (list->len <= eat) {
2171 /* Eaten as whole. */
2176 /* Eaten partially. */
2178 if (skb_shared(list)) {
2179 /* Sucks! We need to fork list. :-( */
2180 clone = skb_clone(list, GFP_ATOMIC);
2186 /* This may be pulled without
2190 if (!pskb_pull(list, eat)) {
2198 /* Free pulled out fragments. */
2199 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2200 skb_shinfo(skb)->frag_list = list->next;
2203 /* And insert new clone at head. */
2206 skb_shinfo(skb)->frag_list = clone;
2209 /* Success! Now we may commit changes to skb data. */
2214 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2215 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2218 skb_frag_unref(skb, i);
2221 skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2223 *frag = skb_shinfo(skb)->frags[i];
2225 skb_frag_off_add(frag, eat);
2226 skb_frag_size_sub(frag, eat);
2234 skb_shinfo(skb)->nr_frags = k;
2238 skb->data_len -= delta;
2241 skb_zcopy_clear(skb, false);
2243 return skb_tail_pointer(skb);
2245 EXPORT_SYMBOL(__pskb_pull_tail);
2248 * skb_copy_bits - copy bits from skb to kernel buffer
2250 * @offset: offset in source
2251 * @to: destination buffer
2252 * @len: number of bytes to copy
2254 * Copy the specified number of bytes from the source skb to the
2255 * destination buffer.
2258 * If its prototype is ever changed,
2259 * check arch/{*}/net/{*}.S files,
2260 * since it is called from BPF assembly code.
2262 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2264 int start = skb_headlen(skb);
2265 struct sk_buff *frag_iter;
2268 if (offset > (int)skb->len - len)
2272 if ((copy = start - offset) > 0) {
2275 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2276 if ((len -= copy) == 0)
2282 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2284 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2286 WARN_ON(start > offset + len);
2288 end = start + skb_frag_size(f);
2289 if ((copy = end - offset) > 0) {
2290 u32 p_off, p_len, copied;
2297 skb_frag_foreach_page(f,
2298 skb_frag_off(f) + offset - start,
2299 copy, p, p_off, p_len, copied) {
2300 vaddr = kmap_atomic(p);
2301 memcpy(to + copied, vaddr + p_off, p_len);
2302 kunmap_atomic(vaddr);
2305 if ((len -= copy) == 0)
2313 skb_walk_frags(skb, frag_iter) {
2316 WARN_ON(start > offset + len);
2318 end = start + frag_iter->len;
2319 if ((copy = end - offset) > 0) {
2322 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2324 if ((len -= copy) == 0)
2338 EXPORT_SYMBOL(skb_copy_bits);
2341 * Callback from splice_to_pipe(), if we need to release some pages
2342 * at the end of the spd in case we error'ed out in filling the pipe.
2344 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2346 put_page(spd->pages[i]);
2349 static struct page *linear_to_page(struct page *page, unsigned int *len,
2350 unsigned int *offset,
2353 struct page_frag *pfrag = sk_page_frag(sk);
2355 if (!sk_page_frag_refill(sk, pfrag))
2358 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2360 memcpy(page_address(pfrag->page) + pfrag->offset,
2361 page_address(page) + *offset, *len);
2362 *offset = pfrag->offset;
2363 pfrag->offset += *len;
2368 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2370 unsigned int offset)
2372 return spd->nr_pages &&
2373 spd->pages[spd->nr_pages - 1] == page &&
2374 (spd->partial[spd->nr_pages - 1].offset +
2375 spd->partial[spd->nr_pages - 1].len == offset);
2379 * Fill page/offset/length into spd, if it can hold more pages.
2381 static bool spd_fill_page(struct splice_pipe_desc *spd,
2382 struct pipe_inode_info *pipe, struct page *page,
2383 unsigned int *len, unsigned int offset,
2387 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2391 page = linear_to_page(page, len, &offset, sk);
2395 if (spd_can_coalesce(spd, page, offset)) {
2396 spd->partial[spd->nr_pages - 1].len += *len;
2400 spd->pages[spd->nr_pages] = page;
2401 spd->partial[spd->nr_pages].len = *len;
2402 spd->partial[spd->nr_pages].offset = offset;
2408 static bool __splice_segment(struct page *page, unsigned int poff,
2409 unsigned int plen, unsigned int *off,
2411 struct splice_pipe_desc *spd, bool linear,
2413 struct pipe_inode_info *pipe)
2418 /* skip this segment if already processed */
2424 /* ignore any bits we already processed */
2430 unsigned int flen = min(*len, plen);
2432 if (spd_fill_page(spd, pipe, page, &flen, poff,
2438 } while (*len && plen);
2444 * Map linear and fragment data from the skb to spd. It reports true if the
2445 * pipe is full or if we already spliced the requested length.
2447 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2448 unsigned int *offset, unsigned int *len,
2449 struct splice_pipe_desc *spd, struct sock *sk)
2452 struct sk_buff *iter;
2454 /* map the linear part :
2455 * If skb->head_frag is set, this 'linear' part is backed by a
2456 * fragment, and if the head is not shared with any clones then
2457 * we can avoid a copy since we own the head portion of this page.
2459 if (__splice_segment(virt_to_page(skb->data),
2460 (unsigned long) skb->data & (PAGE_SIZE - 1),
2463 skb_head_is_locked(skb),
2468 * then map the fragments
2470 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2471 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2473 if (__splice_segment(skb_frag_page(f),
2474 skb_frag_off(f), skb_frag_size(f),
2475 offset, len, spd, false, sk, pipe))
2479 skb_walk_frags(skb, iter) {
2480 if (*offset >= iter->len) {
2481 *offset -= iter->len;
2484 /* __skb_splice_bits() only fails if the output has no room
2485 * left, so no point in going over the frag_list for the error
2488 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2496 * Map data from the skb to a pipe. Should handle both the linear part,
2497 * the fragments, and the frag list.
2499 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2500 struct pipe_inode_info *pipe, unsigned int tlen,
2503 struct partial_page partial[MAX_SKB_FRAGS];
2504 struct page *pages[MAX_SKB_FRAGS];
2505 struct splice_pipe_desc spd = {
2508 .nr_pages_max = MAX_SKB_FRAGS,
2509 .ops = &nosteal_pipe_buf_ops,
2510 .spd_release = sock_spd_release,
2514 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2517 ret = splice_to_pipe(pipe, &spd);
2521 EXPORT_SYMBOL_GPL(skb_splice_bits);
2523 static int sendmsg_unlocked(struct sock *sk, struct msghdr *msg,
2524 struct kvec *vec, size_t num, size_t size)
2526 struct socket *sock = sk->sk_socket;
2530 return kernel_sendmsg(sock, msg, vec, num, size);
2533 static int sendpage_unlocked(struct sock *sk, struct page *page, int offset,
2534 size_t size, int flags)
2536 struct socket *sock = sk->sk_socket;
2540 return kernel_sendpage(sock, page, offset, size, flags);
2543 typedef int (*sendmsg_func)(struct sock *sk, struct msghdr *msg,
2544 struct kvec *vec, size_t num, size_t size);
2545 typedef int (*sendpage_func)(struct sock *sk, struct page *page, int offset,
2546 size_t size, int flags);
2547 static int __skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset,
2548 int len, sendmsg_func sendmsg, sendpage_func sendpage)
2550 unsigned int orig_len = len;
2551 struct sk_buff *head = skb;
2552 unsigned short fragidx;
2557 /* Deal with head data */
2558 while (offset < skb_headlen(skb) && len) {
2562 slen = min_t(int, len, skb_headlen(skb) - offset);
2563 kv.iov_base = skb->data + offset;
2565 memset(&msg, 0, sizeof(msg));
2566 msg.msg_flags = MSG_DONTWAIT;
2568 ret = INDIRECT_CALL_2(sendmsg, kernel_sendmsg_locked,
2569 sendmsg_unlocked, sk, &msg, &kv, 1, slen);
2577 /* All the data was skb head? */
2581 /* Make offset relative to start of frags */
2582 offset -= skb_headlen(skb);
2584 /* Find where we are in frag list */
2585 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2586 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2588 if (offset < skb_frag_size(frag))
2591 offset -= skb_frag_size(frag);
2594 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2595 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2597 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2600 ret = INDIRECT_CALL_2(sendpage, kernel_sendpage_locked,
2601 sendpage_unlocked, sk,
2602 skb_frag_page(frag),
2603 skb_frag_off(frag) + offset,
2604 slen, MSG_DONTWAIT);
2617 /* Process any frag lists */
2620 if (skb_has_frag_list(skb)) {
2621 skb = skb_shinfo(skb)->frag_list;
2624 } else if (skb->next) {
2631 return orig_len - len;
2634 return orig_len == len ? ret : orig_len - len;
2637 /* Send skb data on a socket. Socket must be locked. */
2638 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2641 return __skb_send_sock(sk, skb, offset, len, kernel_sendmsg_locked,
2642 kernel_sendpage_locked);
2644 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2646 /* Send skb data on a socket. Socket must be unlocked. */
2647 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len)
2649 return __skb_send_sock(sk, skb, offset, len, sendmsg_unlocked,
2654 * skb_store_bits - store bits from kernel buffer to skb
2655 * @skb: destination buffer
2656 * @offset: offset in destination
2657 * @from: source buffer
2658 * @len: number of bytes to copy
2660 * Copy the specified number of bytes from the source buffer to the
2661 * destination skb. This function handles all the messy bits of
2662 * traversing fragment lists and such.
2665 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2667 int start = skb_headlen(skb);
2668 struct sk_buff *frag_iter;
2671 if (offset > (int)skb->len - len)
2674 if ((copy = start - offset) > 0) {
2677 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2678 if ((len -= copy) == 0)
2684 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2685 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2688 WARN_ON(start > offset + len);
2690 end = start + skb_frag_size(frag);
2691 if ((copy = end - offset) > 0) {
2692 u32 p_off, p_len, copied;
2699 skb_frag_foreach_page(frag,
2700 skb_frag_off(frag) + offset - start,
2701 copy, p, p_off, p_len, copied) {
2702 vaddr = kmap_atomic(p);
2703 memcpy(vaddr + p_off, from + copied, p_len);
2704 kunmap_atomic(vaddr);
2707 if ((len -= copy) == 0)
2715 skb_walk_frags(skb, frag_iter) {
2718 WARN_ON(start > offset + len);
2720 end = start + frag_iter->len;
2721 if ((copy = end - offset) > 0) {
2724 if (skb_store_bits(frag_iter, offset - start,
2727 if ((len -= copy) == 0)
2740 EXPORT_SYMBOL(skb_store_bits);
2742 /* Checksum skb data. */
2743 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2744 __wsum csum, const struct skb_checksum_ops *ops)
2746 int start = skb_headlen(skb);
2747 int i, copy = start - offset;
2748 struct sk_buff *frag_iter;
2751 /* Checksum header. */
2755 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
2756 skb->data + offset, copy, csum);
2757 if ((len -= copy) == 0)
2763 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2765 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2767 WARN_ON(start > offset + len);
2769 end = start + skb_frag_size(frag);
2770 if ((copy = end - offset) > 0) {
2771 u32 p_off, p_len, copied;
2779 skb_frag_foreach_page(frag,
2780 skb_frag_off(frag) + offset - start,
2781 copy, p, p_off, p_len, copied) {
2782 vaddr = kmap_atomic(p);
2783 csum2 = INDIRECT_CALL_1(ops->update,
2785 vaddr + p_off, p_len, 0);
2786 kunmap_atomic(vaddr);
2787 csum = INDIRECT_CALL_1(ops->combine,
2788 csum_block_add_ext, csum,
2800 skb_walk_frags(skb, frag_iter) {
2803 WARN_ON(start > offset + len);
2805 end = start + frag_iter->len;
2806 if ((copy = end - offset) > 0) {
2810 csum2 = __skb_checksum(frag_iter, offset - start,
2812 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
2813 csum, csum2, pos, copy);
2814 if ((len -= copy) == 0)
2825 EXPORT_SYMBOL(__skb_checksum);
2827 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2828 int len, __wsum csum)
2830 const struct skb_checksum_ops ops = {
2831 .update = csum_partial_ext,
2832 .combine = csum_block_add_ext,
2835 return __skb_checksum(skb, offset, len, csum, &ops);
2837 EXPORT_SYMBOL(skb_checksum);
2839 /* Both of above in one bottle. */
2841 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2844 int start = skb_headlen(skb);
2845 int i, copy = start - offset;
2846 struct sk_buff *frag_iter;
2854 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2856 if ((len -= copy) == 0)
2863 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2866 WARN_ON(start > offset + len);
2868 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2869 if ((copy = end - offset) > 0) {
2870 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2871 u32 p_off, p_len, copied;
2879 skb_frag_foreach_page(frag,
2880 skb_frag_off(frag) + offset - start,
2881 copy, p, p_off, p_len, copied) {
2882 vaddr = kmap_atomic(p);
2883 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2886 kunmap_atomic(vaddr);
2887 csum = csum_block_add(csum, csum2, pos);
2899 skb_walk_frags(skb, frag_iter) {
2903 WARN_ON(start > offset + len);
2905 end = start + frag_iter->len;
2906 if ((copy = end - offset) > 0) {
2909 csum2 = skb_copy_and_csum_bits(frag_iter,
2912 csum = csum_block_add(csum, csum2, pos);
2913 if ((len -= copy) == 0)
2924 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2926 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
2930 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
2931 /* See comments in __skb_checksum_complete(). */
2933 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2934 !skb->csum_complete_sw)
2935 netdev_rx_csum_fault(skb->dev, skb);
2937 if (!skb_shared(skb))
2938 skb->csum_valid = !sum;
2941 EXPORT_SYMBOL(__skb_checksum_complete_head);
2943 /* This function assumes skb->csum already holds pseudo header's checksum,
2944 * which has been changed from the hardware checksum, for example, by
2945 * __skb_checksum_validate_complete(). And, the original skb->csum must
2946 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
2948 * It returns non-zero if the recomputed checksum is still invalid, otherwise
2949 * zero. The new checksum is stored back into skb->csum unless the skb is
2952 __sum16 __skb_checksum_complete(struct sk_buff *skb)
2957 csum = skb_checksum(skb, 0, skb->len, 0);
2959 sum = csum_fold(csum_add(skb->csum, csum));
2960 /* This check is inverted, because we already knew the hardware
2961 * checksum is invalid before calling this function. So, if the
2962 * re-computed checksum is valid instead, then we have a mismatch
2963 * between the original skb->csum and skb_checksum(). This means either
2964 * the original hardware checksum is incorrect or we screw up skb->csum
2965 * when moving skb->data around.
2968 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2969 !skb->csum_complete_sw)
2970 netdev_rx_csum_fault(skb->dev, skb);
2973 if (!skb_shared(skb)) {
2974 /* Save full packet checksum */
2976 skb->ip_summed = CHECKSUM_COMPLETE;
2977 skb->csum_complete_sw = 1;
2978 skb->csum_valid = !sum;
2983 EXPORT_SYMBOL(__skb_checksum_complete);
2985 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2987 net_warn_ratelimited(
2988 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2993 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2994 int offset, int len)
2996 net_warn_ratelimited(
2997 "%s: attempt to compute crc32c without libcrc32c.ko\n",
3002 static const struct skb_checksum_ops default_crc32c_ops = {
3003 .update = warn_crc32c_csum_update,
3004 .combine = warn_crc32c_csum_combine,
3007 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
3008 &default_crc32c_ops;
3009 EXPORT_SYMBOL(crc32c_csum_stub);
3012 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
3013 * @from: source buffer
3015 * Calculates the amount of linear headroom needed in the 'to' skb passed
3016 * into skb_zerocopy().
3019 skb_zerocopy_headlen(const struct sk_buff *from)
3021 unsigned int hlen = 0;
3023 if (!from->head_frag ||
3024 skb_headlen(from) < L1_CACHE_BYTES ||
3025 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) {
3026 hlen = skb_headlen(from);
3031 if (skb_has_frag_list(from))
3036 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
3039 * skb_zerocopy - Zero copy skb to skb
3040 * @to: destination buffer
3041 * @from: source buffer
3042 * @len: number of bytes to copy from source buffer
3043 * @hlen: size of linear headroom in destination buffer
3045 * Copies up to `len` bytes from `from` to `to` by creating references
3046 * to the frags in the source buffer.
3048 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
3049 * headroom in the `to` buffer.
3052 * 0: everything is OK
3053 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
3054 * -EFAULT: skb_copy_bits() found some problem with skb geometry
3057 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
3060 int plen = 0; /* length of skb->head fragment */
3063 unsigned int offset;
3065 BUG_ON(!from->head_frag && !hlen);
3067 /* dont bother with small payloads */
3068 if (len <= skb_tailroom(to))
3069 return skb_copy_bits(from, 0, skb_put(to, len), len);
3072 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
3077 plen = min_t(int, skb_headlen(from), len);
3079 page = virt_to_head_page(from->head);
3080 offset = from->data - (unsigned char *)page_address(page);
3081 __skb_fill_page_desc(to, 0, page, offset, plen);
3088 to->truesize += len + plen;
3089 to->len += len + plen;
3090 to->data_len += len + plen;
3092 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
3096 skb_zerocopy_clone(to, from, GFP_ATOMIC);
3098 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
3103 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
3104 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
3106 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
3108 skb_frag_ref(to, j);
3111 skb_shinfo(to)->nr_frags = j;
3115 EXPORT_SYMBOL_GPL(skb_zerocopy);
3117 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3122 if (skb->ip_summed == CHECKSUM_PARTIAL)
3123 csstart = skb_checksum_start_offset(skb);
3125 csstart = skb_headlen(skb);
3127 BUG_ON(csstart > skb_headlen(skb));
3129 skb_copy_from_linear_data(skb, to, csstart);
3132 if (csstart != skb->len)
3133 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3134 skb->len - csstart);
3136 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3137 long csstuff = csstart + skb->csum_offset;
3139 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3142 EXPORT_SYMBOL(skb_copy_and_csum_dev);
3145 * skb_dequeue - remove from the head of the queue
3146 * @list: list to dequeue from
3148 * Remove the head of the list. The list lock is taken so the function
3149 * may be used safely with other locking list functions. The head item is
3150 * returned or %NULL if the list is empty.
3153 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3155 unsigned long flags;
3156 struct sk_buff *result;
3158 spin_lock_irqsave(&list->lock, flags);
3159 result = __skb_dequeue(list);
3160 spin_unlock_irqrestore(&list->lock, flags);
3163 EXPORT_SYMBOL(skb_dequeue);
3166 * skb_dequeue_tail - remove from the tail of the queue
3167 * @list: list to dequeue from
3169 * Remove the tail of the list. The list lock is taken so the function
3170 * may be used safely with other locking list functions. The tail item is
3171 * returned or %NULL if the list is empty.
3173 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3175 unsigned long flags;
3176 struct sk_buff *result;
3178 spin_lock_irqsave(&list->lock, flags);
3179 result = __skb_dequeue_tail(list);
3180 spin_unlock_irqrestore(&list->lock, flags);
3183 EXPORT_SYMBOL(skb_dequeue_tail);
3186 * skb_queue_purge - empty a list
3187 * @list: list to empty
3189 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3190 * the list and one reference dropped. This function takes the list
3191 * lock and is atomic with respect to other list locking functions.
3193 void skb_queue_purge(struct sk_buff_head *list)
3195 struct sk_buff *skb;
3196 while ((skb = skb_dequeue(list)) != NULL)
3199 EXPORT_SYMBOL(skb_queue_purge);
3202 * skb_rbtree_purge - empty a skb rbtree
3203 * @root: root of the rbtree to empty
3204 * Return value: the sum of truesizes of all purged skbs.
3206 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3207 * the list and one reference dropped. This function does not take
3208 * any lock. Synchronization should be handled by the caller (e.g., TCP
3209 * out-of-order queue is protected by the socket lock).
3211 unsigned int skb_rbtree_purge(struct rb_root *root)
3213 struct rb_node *p = rb_first(root);
3214 unsigned int sum = 0;
3217 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3220 rb_erase(&skb->rbnode, root);
3221 sum += skb->truesize;
3228 * skb_queue_head - queue a buffer at the list head
3229 * @list: list to use
3230 * @newsk: buffer to queue
3232 * Queue a buffer at the start of the list. This function takes the
3233 * list lock and can be used safely with other locking &sk_buff functions
3236 * A buffer cannot be placed on two lists at the same time.
3238 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3240 unsigned long flags;
3242 spin_lock_irqsave(&list->lock, flags);
3243 __skb_queue_head(list, newsk);
3244 spin_unlock_irqrestore(&list->lock, flags);
3246 EXPORT_SYMBOL(skb_queue_head);
3249 * skb_queue_tail - queue a buffer at the list tail
3250 * @list: list to use
3251 * @newsk: buffer to queue
3253 * Queue a buffer at the tail of the list. This function takes the
3254 * list lock and can be used safely with other locking &sk_buff functions
3257 * A buffer cannot be placed on two lists at the same time.
3259 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3261 unsigned long flags;
3263 spin_lock_irqsave(&list->lock, flags);
3264 __skb_queue_tail(list, newsk);
3265 spin_unlock_irqrestore(&list->lock, flags);
3267 EXPORT_SYMBOL(skb_queue_tail);
3270 * skb_unlink - remove a buffer from a list
3271 * @skb: buffer to remove
3272 * @list: list to use
3274 * Remove a packet from a list. The list locks are taken and this
3275 * function is atomic with respect to other list locked calls
3277 * You must know what list the SKB is on.
3279 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3281 unsigned long flags;
3283 spin_lock_irqsave(&list->lock, flags);
3284 __skb_unlink(skb, list);
3285 spin_unlock_irqrestore(&list->lock, flags);
3287 EXPORT_SYMBOL(skb_unlink);
3290 * skb_append - append a buffer
3291 * @old: buffer to insert after
3292 * @newsk: buffer to insert
3293 * @list: list to use
3295 * Place a packet after a given packet in a list. The list locks are taken
3296 * and this function is atomic with respect to other list locked calls.
3297 * A buffer cannot be placed on two lists at the same time.
3299 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3301 unsigned long flags;
3303 spin_lock_irqsave(&list->lock, flags);
3304 __skb_queue_after(list, old, newsk);
3305 spin_unlock_irqrestore(&list->lock, flags);
3307 EXPORT_SYMBOL(skb_append);
3309 static inline void skb_split_inside_header(struct sk_buff *skb,
3310 struct sk_buff* skb1,
3311 const u32 len, const int pos)
3315 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3317 /* And move data appendix as is. */
3318 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3319 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3321 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3322 skb_shinfo(skb)->nr_frags = 0;
3323 skb1->data_len = skb->data_len;
3324 skb1->len += skb1->data_len;
3327 skb_set_tail_pointer(skb, len);
3330 static inline void skb_split_no_header(struct sk_buff *skb,
3331 struct sk_buff* skb1,
3332 const u32 len, int pos)
3335 const int nfrags = skb_shinfo(skb)->nr_frags;
3337 skb_shinfo(skb)->nr_frags = 0;
3338 skb1->len = skb1->data_len = skb->len - len;
3340 skb->data_len = len - pos;
3342 for (i = 0; i < nfrags; i++) {
3343 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3345 if (pos + size > len) {
3346 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3350 * We have two variants in this case:
3351 * 1. Move all the frag to the second
3352 * part, if it is possible. F.e.
3353 * this approach is mandatory for TUX,
3354 * where splitting is expensive.
3355 * 2. Split is accurately. We make this.
3357 skb_frag_ref(skb, i);
3358 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3359 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3360 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3361 skb_shinfo(skb)->nr_frags++;
3365 skb_shinfo(skb)->nr_frags++;
3368 skb_shinfo(skb1)->nr_frags = k;
3372 * skb_split - Split fragmented skb to two parts at length len.
3373 * @skb: the buffer to split
3374 * @skb1: the buffer to receive the second part
3375 * @len: new length for skb
3377 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3379 int pos = skb_headlen(skb);
3381 skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3382 skb_zerocopy_clone(skb1, skb, 0);
3383 if (len < pos) /* Split line is inside header. */
3384 skb_split_inside_header(skb, skb1, len, pos);
3385 else /* Second chunk has no header, nothing to copy. */
3386 skb_split_no_header(skb, skb1, len, pos);
3388 EXPORT_SYMBOL(skb_split);
3390 /* Shifting from/to a cloned skb is a no-go.
3392 * Caller cannot keep skb_shinfo related pointers past calling here!
3394 static int skb_prepare_for_shift(struct sk_buff *skb)
3398 if (skb_cloned(skb)) {
3399 /* Save and restore truesize: pskb_expand_head() may reallocate
3400 * memory where ksize(kmalloc(S)) != ksize(kmalloc(S)), but we
3401 * cannot change truesize at this point.
3403 unsigned int save_truesize = skb->truesize;
3405 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3406 skb->truesize = save_truesize;
3412 * skb_shift - Shifts paged data partially from skb to another
3413 * @tgt: buffer into which tail data gets added
3414 * @skb: buffer from which the paged data comes from
3415 * @shiftlen: shift up to this many bytes
3417 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3418 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3419 * It's up to caller to free skb if everything was shifted.
3421 * If @tgt runs out of frags, the whole operation is aborted.
3423 * Skb cannot include anything else but paged data while tgt is allowed
3424 * to have non-paged data as well.
3426 * TODO: full sized shift could be optimized but that would need
3427 * specialized skb free'er to handle frags without up-to-date nr_frags.
3429 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3431 int from, to, merge, todo;
3432 skb_frag_t *fragfrom, *fragto;
3434 BUG_ON(shiftlen > skb->len);
3436 if (skb_headlen(skb))
3438 if (skb_zcopy(tgt) || skb_zcopy(skb))
3443 to = skb_shinfo(tgt)->nr_frags;
3444 fragfrom = &skb_shinfo(skb)->frags[from];
3446 /* Actual merge is delayed until the point when we know we can
3447 * commit all, so that we don't have to undo partial changes
3450 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3451 skb_frag_off(fragfrom))) {
3456 todo -= skb_frag_size(fragfrom);
3458 if (skb_prepare_for_shift(skb) ||
3459 skb_prepare_for_shift(tgt))
3462 /* All previous frag pointers might be stale! */
3463 fragfrom = &skb_shinfo(skb)->frags[from];
3464 fragto = &skb_shinfo(tgt)->frags[merge];
3466 skb_frag_size_add(fragto, shiftlen);
3467 skb_frag_size_sub(fragfrom, shiftlen);
3468 skb_frag_off_add(fragfrom, shiftlen);
3476 /* Skip full, not-fitting skb to avoid expensive operations */
3477 if ((shiftlen == skb->len) &&
3478 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3481 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3484 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3485 if (to == MAX_SKB_FRAGS)
3488 fragfrom = &skb_shinfo(skb)->frags[from];
3489 fragto = &skb_shinfo(tgt)->frags[to];
3491 if (todo >= skb_frag_size(fragfrom)) {
3492 *fragto = *fragfrom;
3493 todo -= skb_frag_size(fragfrom);
3498 __skb_frag_ref(fragfrom);
3499 skb_frag_page_copy(fragto, fragfrom);
3500 skb_frag_off_copy(fragto, fragfrom);
3501 skb_frag_size_set(fragto, todo);
3503 skb_frag_off_add(fragfrom, todo);
3504 skb_frag_size_sub(fragfrom, todo);
3512 /* Ready to "commit" this state change to tgt */
3513 skb_shinfo(tgt)->nr_frags = to;
3516 fragfrom = &skb_shinfo(skb)->frags[0];
3517 fragto = &skb_shinfo(tgt)->frags[merge];
3519 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3520 __skb_frag_unref(fragfrom, skb->pp_recycle);
3523 /* Reposition in the original skb */
3525 while (from < skb_shinfo(skb)->nr_frags)
3526 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3527 skb_shinfo(skb)->nr_frags = to;
3529 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3532 /* Most likely the tgt won't ever need its checksum anymore, skb on
3533 * the other hand might need it if it needs to be resent
3535 tgt->ip_summed = CHECKSUM_PARTIAL;
3536 skb->ip_summed = CHECKSUM_PARTIAL;
3538 /* Yak, is it really working this way? Some helper please? */
3539 skb->len -= shiftlen;
3540 skb->data_len -= shiftlen;
3541 skb->truesize -= shiftlen;
3542 tgt->len += shiftlen;
3543 tgt->data_len += shiftlen;
3544 tgt->truesize += shiftlen;
3550 * skb_prepare_seq_read - Prepare a sequential read of skb data
3551 * @skb: the buffer to read
3552 * @from: lower offset of data to be read
3553 * @to: upper offset of data to be read
3554 * @st: state variable
3556 * Initializes the specified state variable. Must be called before
3557 * invoking skb_seq_read() for the first time.
3559 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3560 unsigned int to, struct skb_seq_state *st)
3562 st->lower_offset = from;
3563 st->upper_offset = to;
3564 st->root_skb = st->cur_skb = skb;
3565 st->frag_idx = st->stepped_offset = 0;
3566 st->frag_data = NULL;
3569 EXPORT_SYMBOL(skb_prepare_seq_read);
3572 * skb_seq_read - Sequentially read skb data
3573 * @consumed: number of bytes consumed by the caller so far
3574 * @data: destination pointer for data to be returned
3575 * @st: state variable
3577 * Reads a block of skb data at @consumed relative to the
3578 * lower offset specified to skb_prepare_seq_read(). Assigns
3579 * the head of the data block to @data and returns the length
3580 * of the block or 0 if the end of the skb data or the upper
3581 * offset has been reached.
3583 * The caller is not required to consume all of the data
3584 * returned, i.e. @consumed is typically set to the number
3585 * of bytes already consumed and the next call to
3586 * skb_seq_read() will return the remaining part of the block.
3588 * Note 1: The size of each block of data returned can be arbitrary,
3589 * this limitation is the cost for zerocopy sequential
3590 * reads of potentially non linear data.
3592 * Note 2: Fragment lists within fragments are not implemented
3593 * at the moment, state->root_skb could be replaced with
3594 * a stack for this purpose.
3596 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3597 struct skb_seq_state *st)
3599 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3602 if (unlikely(abs_offset >= st->upper_offset)) {
3603 if (st->frag_data) {
3604 kunmap_atomic(st->frag_data);
3605 st->frag_data = NULL;
3611 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3613 if (abs_offset < block_limit && !st->frag_data) {
3614 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3615 return block_limit - abs_offset;
3618 if (st->frag_idx == 0 && !st->frag_data)
3619 st->stepped_offset += skb_headlen(st->cur_skb);
3621 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3622 unsigned int pg_idx, pg_off, pg_sz;
3624 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3627 pg_off = skb_frag_off(frag);
3628 pg_sz = skb_frag_size(frag);
3630 if (skb_frag_must_loop(skb_frag_page(frag))) {
3631 pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT;
3632 pg_off = offset_in_page(pg_off + st->frag_off);
3633 pg_sz = min_t(unsigned int, pg_sz - st->frag_off,
3634 PAGE_SIZE - pg_off);
3637 block_limit = pg_sz + st->stepped_offset;
3638 if (abs_offset < block_limit) {
3640 st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx);
3642 *data = (u8 *)st->frag_data + pg_off +
3643 (abs_offset - st->stepped_offset);
3645 return block_limit - abs_offset;
3648 if (st->frag_data) {
3649 kunmap_atomic(st->frag_data);
3650 st->frag_data = NULL;
3653 st->stepped_offset += pg_sz;
3654 st->frag_off += pg_sz;
3655 if (st->frag_off == skb_frag_size(frag)) {
3661 if (st->frag_data) {
3662 kunmap_atomic(st->frag_data);
3663 st->frag_data = NULL;
3666 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3667 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3670 } else if (st->cur_skb->next) {
3671 st->cur_skb = st->cur_skb->next;
3678 EXPORT_SYMBOL(skb_seq_read);
3681 * skb_abort_seq_read - Abort a sequential read of skb data
3682 * @st: state variable
3684 * Must be called if skb_seq_read() was not called until it
3687 void skb_abort_seq_read(struct skb_seq_state *st)
3690 kunmap_atomic(st->frag_data);
3692 EXPORT_SYMBOL(skb_abort_seq_read);
3694 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3696 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3697 struct ts_config *conf,
3698 struct ts_state *state)
3700 return skb_seq_read(offset, text, TS_SKB_CB(state));
3703 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3705 skb_abort_seq_read(TS_SKB_CB(state));
3709 * skb_find_text - Find a text pattern in skb data
3710 * @skb: the buffer to look in
3711 * @from: search offset
3713 * @config: textsearch configuration
3715 * Finds a pattern in the skb data according to the specified
3716 * textsearch configuration. Use textsearch_next() to retrieve
3717 * subsequent occurrences of the pattern. Returns the offset
3718 * to the first occurrence or UINT_MAX if no match was found.
3720 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3721 unsigned int to, struct ts_config *config)
3723 struct ts_state state;
3726 BUILD_BUG_ON(sizeof(struct skb_seq_state) > sizeof(state.cb));
3728 config->get_next_block = skb_ts_get_next_block;
3729 config->finish = skb_ts_finish;
3731 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3733 ret = textsearch_find(config, &state);
3734 return (ret <= to - from ? ret : UINT_MAX);
3736 EXPORT_SYMBOL(skb_find_text);
3738 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3739 int offset, size_t size)
3741 int i = skb_shinfo(skb)->nr_frags;
3743 if (skb_can_coalesce(skb, i, page, offset)) {
3744 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3745 } else if (i < MAX_SKB_FRAGS) {
3747 skb_fill_page_desc(skb, i, page, offset, size);
3754 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3757 * skb_pull_rcsum - pull skb and update receive checksum
3758 * @skb: buffer to update
3759 * @len: length of data pulled
3761 * This function performs an skb_pull on the packet and updates
3762 * the CHECKSUM_COMPLETE checksum. It should be used on
3763 * receive path processing instead of skb_pull unless you know
3764 * that the checksum difference is zero (e.g., a valid IP header)
3765 * or you are setting ip_summed to CHECKSUM_NONE.
3767 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3769 unsigned char *data = skb->data;
3771 BUG_ON(len > skb->len);
3772 __skb_pull(skb, len);
3773 skb_postpull_rcsum(skb, data, len);
3776 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3778 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3780 skb_frag_t head_frag;
3783 page = virt_to_head_page(frag_skb->head);
3784 __skb_frag_set_page(&head_frag, page);
3785 skb_frag_off_set(&head_frag, frag_skb->data -
3786 (unsigned char *)page_address(page));
3787 skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
3791 struct sk_buff *skb_segment_list(struct sk_buff *skb,
3792 netdev_features_t features,
3793 unsigned int offset)
3795 struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
3796 unsigned int tnl_hlen = skb_tnl_header_len(skb);
3797 unsigned int delta_truesize = 0;
3798 unsigned int delta_len = 0;
3799 struct sk_buff *tail = NULL;
3800 struct sk_buff *nskb, *tmp;
3803 skb_push(skb, -skb_network_offset(skb) + offset);
3805 skb_shinfo(skb)->frag_list = NULL;
3809 list_skb = list_skb->next;
3812 if (skb_shared(nskb)) {
3813 tmp = skb_clone(nskb, GFP_ATOMIC);
3817 err = skb_unclone(nskb, GFP_ATOMIC);
3828 if (unlikely(err)) {
3829 nskb->next = list_skb;
3835 delta_len += nskb->len;
3836 delta_truesize += nskb->truesize;
3838 skb_push(nskb, -skb_network_offset(nskb) + offset);
3840 skb_release_head_state(nskb);
3841 __copy_skb_header(nskb, skb);
3843 skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
3844 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
3845 nskb->data - tnl_hlen,
3848 if (skb_needs_linearize(nskb, features) &&
3849 __skb_linearize(nskb))
3854 skb->truesize = skb->truesize - delta_truesize;
3855 skb->data_len = skb->data_len - delta_len;
3856 skb->len = skb->len - delta_len;
3862 if (skb_needs_linearize(skb, features) &&
3863 __skb_linearize(skb))
3871 kfree_skb_list(skb->next);
3873 return ERR_PTR(-ENOMEM);
3875 EXPORT_SYMBOL_GPL(skb_segment_list);
3877 int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
3879 if (unlikely(p->len + skb->len >= 65536))
3882 if (NAPI_GRO_CB(p)->last == p)
3883 skb_shinfo(p)->frag_list = skb;
3885 NAPI_GRO_CB(p)->last->next = skb;
3887 skb_pull(skb, skb_gro_offset(skb));
3889 NAPI_GRO_CB(p)->last = skb;
3890 NAPI_GRO_CB(p)->count++;
3891 p->data_len += skb->len;
3892 p->truesize += skb->truesize;
3895 NAPI_GRO_CB(skb)->same_flow = 1;
3901 * skb_segment - Perform protocol segmentation on skb.
3902 * @head_skb: buffer to segment
3903 * @features: features for the output path (see dev->features)
3905 * This function performs segmentation on the given skb. It returns
3906 * a pointer to the first in a list of new skbs for the segments.
3907 * In case of error it returns ERR_PTR(err).
3909 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3910 netdev_features_t features)
3912 struct sk_buff *segs = NULL;
3913 struct sk_buff *tail = NULL;
3914 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3915 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3916 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3917 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3918 struct sk_buff *frag_skb = head_skb;
3919 unsigned int offset = doffset;
3920 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3921 unsigned int partial_segs = 0;
3922 unsigned int headroom;
3923 unsigned int len = head_skb->len;
3926 int nfrags = skb_shinfo(head_skb)->nr_frags;
3931 if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) &&
3932 (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) {
3933 /* gso_size is untrusted, and we have a frag_list with a linear
3934 * non head_frag head.
3936 * (we assume checking the first list_skb member suffices;
3937 * i.e if either of the list_skb members have non head_frag
3938 * head, then the first one has too).
3940 * If head_skb's headlen does not fit requested gso_size, it
3941 * means that the frag_list members do NOT terminate on exact
3942 * gso_size boundaries. Hence we cannot perform skb_frag_t page
3943 * sharing. Therefore we must fallback to copying the frag_list
3944 * skbs; we do so by disabling SG.
3946 if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb))
3947 features &= ~NETIF_F_SG;
3950 __skb_push(head_skb, doffset);
3951 proto = skb_network_protocol(head_skb, NULL);
3952 if (unlikely(!proto))
3953 return ERR_PTR(-EINVAL);
3955 sg = !!(features & NETIF_F_SG);
3956 csum = !!can_checksum_protocol(features, proto);
3958 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3959 if (!(features & NETIF_F_GSO_PARTIAL)) {
3960 struct sk_buff *iter;
3961 unsigned int frag_len;
3964 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3967 /* If we get here then all the required
3968 * GSO features except frag_list are supported.
3969 * Try to split the SKB to multiple GSO SKBs
3970 * with no frag_list.
3971 * Currently we can do that only when the buffers don't
3972 * have a linear part and all the buffers except
3973 * the last are of the same length.
3975 frag_len = list_skb->len;
3976 skb_walk_frags(head_skb, iter) {
3977 if (frag_len != iter->len && iter->next)
3979 if (skb_headlen(iter) && !iter->head_frag)
3985 if (len != frag_len)
3989 /* GSO partial only requires that we trim off any excess that
3990 * doesn't fit into an MSS sized block, so take care of that
3993 partial_segs = len / mss;
3994 if (partial_segs > 1)
3995 mss *= partial_segs;
4001 headroom = skb_headroom(head_skb);
4002 pos = skb_headlen(head_skb);
4005 struct sk_buff *nskb;
4006 skb_frag_t *nskb_frag;
4010 if (unlikely(mss == GSO_BY_FRAGS)) {
4011 len = list_skb->len;
4013 len = head_skb->len - offset;
4018 hsize = skb_headlen(head_skb) - offset;
4020 if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) &&
4021 (skb_headlen(list_skb) == len || sg)) {
4022 BUG_ON(skb_headlen(list_skb) > len);
4025 nfrags = skb_shinfo(list_skb)->nr_frags;
4026 frag = skb_shinfo(list_skb)->frags;
4027 frag_skb = list_skb;
4028 pos += skb_headlen(list_skb);
4030 while (pos < offset + len) {
4031 BUG_ON(i >= nfrags);
4033 size = skb_frag_size(frag);
4034 if (pos + size > offset + len)
4042 nskb = skb_clone(list_skb, GFP_ATOMIC);
4043 list_skb = list_skb->next;
4045 if (unlikely(!nskb))
4048 if (unlikely(pskb_trim(nskb, len))) {
4053 hsize = skb_end_offset(nskb);
4054 if (skb_cow_head(nskb, doffset + headroom)) {
4059 nskb->truesize += skb_end_offset(nskb) - hsize;
4060 skb_release_head_state(nskb);
4061 __skb_push(nskb, doffset);
4065 if (hsize > len || !sg)
4068 nskb = __alloc_skb(hsize + doffset + headroom,
4069 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
4072 if (unlikely(!nskb))
4075 skb_reserve(nskb, headroom);
4076 __skb_put(nskb, doffset);
4085 __copy_skb_header(nskb, head_skb);
4087 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
4088 skb_reset_mac_len(nskb);
4090 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
4091 nskb->data - tnl_hlen,
4092 doffset + tnl_hlen);
4094 if (nskb->len == len + doffset)
4095 goto perform_csum_check;
4099 if (!nskb->remcsum_offload)
4100 nskb->ip_summed = CHECKSUM_NONE;
4101 SKB_GSO_CB(nskb)->csum =
4102 skb_copy_and_csum_bits(head_skb, offset,
4106 SKB_GSO_CB(nskb)->csum_start =
4107 skb_headroom(nskb) + doffset;
4109 skb_copy_bits(head_skb, offset,
4116 nskb_frag = skb_shinfo(nskb)->frags;
4118 skb_copy_from_linear_data_offset(head_skb, offset,
4119 skb_put(nskb, hsize), hsize);
4121 skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags &
4124 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4125 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
4128 while (pos < offset + len) {
4131 nfrags = skb_shinfo(list_skb)->nr_frags;
4132 frag = skb_shinfo(list_skb)->frags;
4133 frag_skb = list_skb;
4134 if (!skb_headlen(list_skb)) {
4137 BUG_ON(!list_skb->head_frag);
4139 /* to make room for head_frag. */
4143 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4144 skb_zerocopy_clone(nskb, frag_skb,
4148 list_skb = list_skb->next;
4151 if (unlikely(skb_shinfo(nskb)->nr_frags >=
4153 net_warn_ratelimited(
4154 "skb_segment: too many frags: %u %u\n",
4160 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
4161 __skb_frag_ref(nskb_frag);
4162 size = skb_frag_size(nskb_frag);
4165 skb_frag_off_add(nskb_frag, offset - pos);
4166 skb_frag_size_sub(nskb_frag, offset - pos);
4169 skb_shinfo(nskb)->nr_frags++;
4171 if (pos + size <= offset + len) {
4176 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4184 nskb->data_len = len - hsize;
4185 nskb->len += nskb->data_len;
4186 nskb->truesize += nskb->data_len;
4190 if (skb_has_shared_frag(nskb) &&
4191 __skb_linearize(nskb))
4194 if (!nskb->remcsum_offload)
4195 nskb->ip_summed = CHECKSUM_NONE;
4196 SKB_GSO_CB(nskb)->csum =
4197 skb_checksum(nskb, doffset,
4198 nskb->len - doffset, 0);
4199 SKB_GSO_CB(nskb)->csum_start =
4200 skb_headroom(nskb) + doffset;
4202 } while ((offset += len) < head_skb->len);
4204 /* Some callers want to get the end of the list.
4205 * Put it in segs->prev to avoid walking the list.
4206 * (see validate_xmit_skb_list() for example)
4211 struct sk_buff *iter;
4212 int type = skb_shinfo(head_skb)->gso_type;
4213 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
4215 /* Update type to add partial and then remove dodgy if set */
4216 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
4217 type &= ~SKB_GSO_DODGY;
4219 /* Update GSO info and prepare to start updating headers on
4220 * our way back down the stack of protocols.
4222 for (iter = segs; iter; iter = iter->next) {
4223 skb_shinfo(iter)->gso_size = gso_size;
4224 skb_shinfo(iter)->gso_segs = partial_segs;
4225 skb_shinfo(iter)->gso_type = type;
4226 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4229 if (tail->len - doffset <= gso_size)
4230 skb_shinfo(tail)->gso_size = 0;
4231 else if (tail != segs)
4232 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4235 /* Following permits correct backpressure, for protocols
4236 * using skb_set_owner_w().
4237 * Idea is to tranfert ownership from head_skb to last segment.
4239 if (head_skb->destructor == sock_wfree) {
4240 swap(tail->truesize, head_skb->truesize);
4241 swap(tail->destructor, head_skb->destructor);
4242 swap(tail->sk, head_skb->sk);
4247 kfree_skb_list(segs);
4248 return ERR_PTR(err);
4250 EXPORT_SYMBOL_GPL(skb_segment);
4252 int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
4254 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
4255 unsigned int offset = skb_gro_offset(skb);
4256 unsigned int headlen = skb_headlen(skb);
4257 unsigned int len = skb_gro_len(skb);
4258 unsigned int delta_truesize;
4261 if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush))
4264 lp = NAPI_GRO_CB(p)->last;
4265 pinfo = skb_shinfo(lp);
4267 if (headlen <= offset) {
4270 int i = skbinfo->nr_frags;
4271 int nr_frags = pinfo->nr_frags + i;
4273 if (nr_frags > MAX_SKB_FRAGS)
4277 pinfo->nr_frags = nr_frags;
4278 skbinfo->nr_frags = 0;
4280 frag = pinfo->frags + nr_frags;
4281 frag2 = skbinfo->frags + i;
4286 skb_frag_off_add(frag, offset);
4287 skb_frag_size_sub(frag, offset);
4289 /* all fragments truesize : remove (head size + sk_buff) */
4290 delta_truesize = skb->truesize -
4291 SKB_TRUESIZE(skb_end_offset(skb));
4293 skb->truesize -= skb->data_len;
4294 skb->len -= skb->data_len;
4297 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
4299 } else if (skb->head_frag) {
4300 int nr_frags = pinfo->nr_frags;
4301 skb_frag_t *frag = pinfo->frags + nr_frags;
4302 struct page *page = virt_to_head_page(skb->head);
4303 unsigned int first_size = headlen - offset;
4304 unsigned int first_offset;
4306 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
4309 first_offset = skb->data -
4310 (unsigned char *)page_address(page) +
4313 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
4315 __skb_frag_set_page(frag, page);
4316 skb_frag_off_set(frag, first_offset);
4317 skb_frag_size_set(frag, first_size);
4319 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
4320 /* We dont need to clear skbinfo->nr_frags here */
4322 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4323 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
4328 delta_truesize = skb->truesize;
4329 if (offset > headlen) {
4330 unsigned int eat = offset - headlen;
4332 skb_frag_off_add(&skbinfo->frags[0], eat);
4333 skb_frag_size_sub(&skbinfo->frags[0], eat);
4334 skb->data_len -= eat;
4339 __skb_pull(skb, offset);
4341 if (NAPI_GRO_CB(p)->last == p)
4342 skb_shinfo(p)->frag_list = skb;
4344 NAPI_GRO_CB(p)->last->next = skb;
4345 NAPI_GRO_CB(p)->last = skb;
4346 __skb_header_release(skb);
4350 NAPI_GRO_CB(p)->count++;
4352 p->truesize += delta_truesize;
4355 lp->data_len += len;
4356 lp->truesize += delta_truesize;
4359 NAPI_GRO_CB(skb)->same_flow = 1;
4363 #ifdef CONFIG_SKB_EXTENSIONS
4364 #define SKB_EXT_ALIGN_VALUE 8
4365 #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4367 static const u8 skb_ext_type_len[] = {
4368 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4369 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4372 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4374 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4375 [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4377 #if IS_ENABLED(CONFIG_MPTCP)
4378 [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
4382 static __always_inline unsigned int skb_ext_total_length(void)
4384 return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4385 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4386 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4389 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4391 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4392 skb_ext_type_len[TC_SKB_EXT] +
4394 #if IS_ENABLED(CONFIG_MPTCP)
4395 skb_ext_type_len[SKB_EXT_MPTCP] +
4400 static void skb_extensions_init(void)
4402 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4403 BUILD_BUG_ON(skb_ext_total_length() > 255);
4405 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4406 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4408 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4412 static void skb_extensions_init(void) {}
4415 void __init skb_init(void)
4417 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4418 sizeof(struct sk_buff),
4420 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4421 offsetof(struct sk_buff, cb),
4422 sizeof_field(struct sk_buff, cb),
4424 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4425 sizeof(struct sk_buff_fclones),
4427 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4429 skb_extensions_init();
4433 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4434 unsigned int recursion_level)
4436 int start = skb_headlen(skb);
4437 int i, copy = start - offset;
4438 struct sk_buff *frag_iter;
4441 if (unlikely(recursion_level >= 24))
4447 sg_set_buf(sg, skb->data + offset, copy);
4449 if ((len -= copy) == 0)
4454 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4457 WARN_ON(start > offset + len);
4459 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4460 if ((copy = end - offset) > 0) {
4461 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4462 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4467 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4468 skb_frag_off(frag) + offset - start);
4477 skb_walk_frags(skb, frag_iter) {
4480 WARN_ON(start > offset + len);
4482 end = start + frag_iter->len;
4483 if ((copy = end - offset) > 0) {
4484 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4489 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4490 copy, recursion_level + 1);
4491 if (unlikely(ret < 0))
4494 if ((len -= copy) == 0)
4505 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4506 * @skb: Socket buffer containing the buffers to be mapped
4507 * @sg: The scatter-gather list to map into
4508 * @offset: The offset into the buffer's contents to start mapping
4509 * @len: Length of buffer space to be mapped
4511 * Fill the specified scatter-gather list with mappings/pointers into a
4512 * region of the buffer space attached to a socket buffer. Returns either
4513 * the number of scatterlist items used, or -EMSGSIZE if the contents
4516 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4518 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4523 sg_mark_end(&sg[nsg - 1]);
4527 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4529 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4530 * sglist without mark the sg which contain last skb data as the end.
4531 * So the caller can mannipulate sg list as will when padding new data after
4532 * the first call without calling sg_unmark_end to expend sg list.
4534 * Scenario to use skb_to_sgvec_nomark:
4536 * 2. skb_to_sgvec_nomark(payload1)
4537 * 3. skb_to_sgvec_nomark(payload2)
4539 * This is equivalent to:
4541 * 2. skb_to_sgvec(payload1)
4543 * 4. skb_to_sgvec(payload2)
4545 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4546 * is more preferable.
4548 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4549 int offset, int len)
4551 return __skb_to_sgvec(skb, sg, offset, len, 0);
4553 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4558 * skb_cow_data - Check that a socket buffer's data buffers are writable
4559 * @skb: The socket buffer to check.
4560 * @tailbits: Amount of trailing space to be added
4561 * @trailer: Returned pointer to the skb where the @tailbits space begins
4563 * Make sure that the data buffers attached to a socket buffer are
4564 * writable. If they are not, private copies are made of the data buffers
4565 * and the socket buffer is set to use these instead.
4567 * If @tailbits is given, make sure that there is space to write @tailbits
4568 * bytes of data beyond current end of socket buffer. @trailer will be
4569 * set to point to the skb in which this space begins.
4571 * The number of scatterlist elements required to completely map the
4572 * COW'd and extended socket buffer will be returned.
4574 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4578 struct sk_buff *skb1, **skb_p;
4580 /* If skb is cloned or its head is paged, reallocate
4581 * head pulling out all the pages (pages are considered not writable
4582 * at the moment even if they are anonymous).
4584 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4585 !__pskb_pull_tail(skb, __skb_pagelen(skb)))
4588 /* Easy case. Most of packets will go this way. */
4589 if (!skb_has_frag_list(skb)) {
4590 /* A little of trouble, not enough of space for trailer.
4591 * This should not happen, when stack is tuned to generate
4592 * good frames. OK, on miss we reallocate and reserve even more
4593 * space, 128 bytes is fair. */
4595 if (skb_tailroom(skb) < tailbits &&
4596 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4604 /* Misery. We are in troubles, going to mincer fragments... */
4607 skb_p = &skb_shinfo(skb)->frag_list;
4610 while ((skb1 = *skb_p) != NULL) {
4613 /* The fragment is partially pulled by someone,
4614 * this can happen on input. Copy it and everything
4617 if (skb_shared(skb1))
4620 /* If the skb is the last, worry about trailer. */
4622 if (skb1->next == NULL && tailbits) {
4623 if (skb_shinfo(skb1)->nr_frags ||
4624 skb_has_frag_list(skb1) ||
4625 skb_tailroom(skb1) < tailbits)
4626 ntail = tailbits + 128;
4632 skb_shinfo(skb1)->nr_frags ||
4633 skb_has_frag_list(skb1)) {
4634 struct sk_buff *skb2;
4636 /* Fuck, we are miserable poor guys... */
4638 skb2 = skb_copy(skb1, GFP_ATOMIC);
4640 skb2 = skb_copy_expand(skb1,
4644 if (unlikely(skb2 == NULL))
4648 skb_set_owner_w(skb2, skb1->sk);
4650 /* Looking around. Are we still alive?
4651 * OK, link new skb, drop old one */
4653 skb2->next = skb1->next;
4660 skb_p = &skb1->next;
4665 EXPORT_SYMBOL_GPL(skb_cow_data);
4667 static void sock_rmem_free(struct sk_buff *skb)
4669 struct sock *sk = skb->sk;
4671 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4674 static void skb_set_err_queue(struct sk_buff *skb)
4676 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4677 * So, it is safe to (mis)use it to mark skbs on the error queue.
4679 skb->pkt_type = PACKET_OUTGOING;
4680 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4684 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4686 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4688 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4689 (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4694 skb->destructor = sock_rmem_free;
4695 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4696 skb_set_err_queue(skb);
4698 /* before exiting rcu section, make sure dst is refcounted */
4701 skb_queue_tail(&sk->sk_error_queue, skb);
4702 if (!sock_flag(sk, SOCK_DEAD))
4703 sk_error_report(sk);
4706 EXPORT_SYMBOL(sock_queue_err_skb);
4708 static bool is_icmp_err_skb(const struct sk_buff *skb)
4710 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4711 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4714 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4716 struct sk_buff_head *q = &sk->sk_error_queue;
4717 struct sk_buff *skb, *skb_next = NULL;
4718 bool icmp_next = false;
4719 unsigned long flags;
4721 spin_lock_irqsave(&q->lock, flags);
4722 skb = __skb_dequeue(q);
4723 if (skb && (skb_next = skb_peek(q))) {
4724 icmp_next = is_icmp_err_skb(skb_next);
4726 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
4728 spin_unlock_irqrestore(&q->lock, flags);
4730 if (is_icmp_err_skb(skb) && !icmp_next)
4734 sk_error_report(sk);
4738 EXPORT_SYMBOL(sock_dequeue_err_skb);
4741 * skb_clone_sk - create clone of skb, and take reference to socket
4742 * @skb: the skb to clone
4744 * This function creates a clone of a buffer that holds a reference on
4745 * sk_refcnt. Buffers created via this function are meant to be
4746 * returned using sock_queue_err_skb, or free via kfree_skb.
4748 * When passing buffers allocated with this function to sock_queue_err_skb
4749 * it is necessary to wrap the call with sock_hold/sock_put in order to
4750 * prevent the socket from being released prior to being enqueued on
4751 * the sk_error_queue.
4753 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4755 struct sock *sk = skb->sk;
4756 struct sk_buff *clone;
4758 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4761 clone = skb_clone(skb, GFP_ATOMIC);
4768 clone->destructor = sock_efree;
4772 EXPORT_SYMBOL(skb_clone_sk);
4774 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4779 struct sock_exterr_skb *serr;
4782 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4784 serr = SKB_EXT_ERR(skb);
4785 memset(serr, 0, sizeof(*serr));
4786 serr->ee.ee_errno = ENOMSG;
4787 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4788 serr->ee.ee_info = tstype;
4789 serr->opt_stats = opt_stats;
4790 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4791 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4792 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4793 if (sk->sk_protocol == IPPROTO_TCP &&
4794 sk->sk_type == SOCK_STREAM)
4795 serr->ee.ee_data -= sk->sk_tskey;
4798 err = sock_queue_err_skb(sk, skb);
4804 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4808 if (likely(sysctl_tstamp_allow_data || tsonly))
4811 read_lock_bh(&sk->sk_callback_lock);
4812 ret = sk->sk_socket && sk->sk_socket->file &&
4813 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4814 read_unlock_bh(&sk->sk_callback_lock);
4818 void skb_complete_tx_timestamp(struct sk_buff *skb,
4819 struct skb_shared_hwtstamps *hwtstamps)
4821 struct sock *sk = skb->sk;
4823 if (!skb_may_tx_timestamp(sk, false))
4826 /* Take a reference to prevent skb_orphan() from freeing the socket,
4827 * but only if the socket refcount is not zero.
4829 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4830 *skb_hwtstamps(skb) = *hwtstamps;
4831 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4839 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4841 void __skb_tstamp_tx(struct sk_buff *orig_skb,
4842 const struct sk_buff *ack_skb,
4843 struct skb_shared_hwtstamps *hwtstamps,
4844 struct sock *sk, int tstype)
4846 struct sk_buff *skb;
4847 bool tsonly, opt_stats = false;
4852 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4853 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4856 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4857 if (!skb_may_tx_timestamp(sk, tsonly))
4862 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4863 sk->sk_protocol == IPPROTO_TCP &&
4864 sk->sk_type == SOCK_STREAM) {
4865 skb = tcp_get_timestamping_opt_stats(sk, orig_skb,
4870 skb = alloc_skb(0, GFP_ATOMIC);
4872 skb = skb_clone(orig_skb, GFP_ATOMIC);
4878 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4880 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4884 *skb_hwtstamps(skb) = *hwtstamps;
4886 skb->tstamp = ktime_get_real();
4888 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4890 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4892 void skb_tstamp_tx(struct sk_buff *orig_skb,
4893 struct skb_shared_hwtstamps *hwtstamps)
4895 return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk,
4898 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4900 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4902 struct sock *sk = skb->sk;
4903 struct sock_exterr_skb *serr;
4906 skb->wifi_acked_valid = 1;
4907 skb->wifi_acked = acked;
4909 serr = SKB_EXT_ERR(skb);
4910 memset(serr, 0, sizeof(*serr));
4911 serr->ee.ee_errno = ENOMSG;
4912 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4914 /* Take a reference to prevent skb_orphan() from freeing the socket,
4915 * but only if the socket refcount is not zero.
4917 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4918 err = sock_queue_err_skb(sk, skb);
4924 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4927 * skb_partial_csum_set - set up and verify partial csum values for packet
4928 * @skb: the skb to set
4929 * @start: the number of bytes after skb->data to start checksumming.
4930 * @off: the offset from start to place the checksum.
4932 * For untrusted partially-checksummed packets, we need to make sure the values
4933 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4935 * This function checks and sets those values and skb->ip_summed: if this
4936 * returns false you should drop the packet.
4938 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4940 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
4941 u32 csum_start = skb_headroom(skb) + (u32)start;
4943 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
4944 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4945 start, off, skb_headroom(skb), skb_headlen(skb));
4948 skb->ip_summed = CHECKSUM_PARTIAL;
4949 skb->csum_start = csum_start;
4950 skb->csum_offset = off;
4951 skb_set_transport_header(skb, start);
4954 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4956 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4959 if (skb_headlen(skb) >= len)
4962 /* If we need to pullup then pullup to the max, so we
4963 * won't need to do it again.
4968 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4971 if (skb_headlen(skb) < len)
4977 #define MAX_TCP_HDR_LEN (15 * 4)
4979 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4980 typeof(IPPROTO_IP) proto,
4987 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4988 off + MAX_TCP_HDR_LEN);
4989 if (!err && !skb_partial_csum_set(skb, off,
4990 offsetof(struct tcphdr,
4993 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4996 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4997 off + sizeof(struct udphdr));
4998 if (!err && !skb_partial_csum_set(skb, off,
4999 offsetof(struct udphdr,
5002 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
5005 return ERR_PTR(-EPROTO);
5008 /* This value should be large enough to cover a tagged ethernet header plus
5009 * maximally sized IP and TCP or UDP headers.
5011 #define MAX_IP_HDR_LEN 128
5013 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
5022 err = skb_maybe_pull_tail(skb,
5023 sizeof(struct iphdr),
5028 if (ip_is_fragment(ip_hdr(skb)))
5031 off = ip_hdrlen(skb);
5038 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
5040 return PTR_ERR(csum);
5043 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
5046 ip_hdr(skb)->protocol, 0);
5053 /* This value should be large enough to cover a tagged ethernet header plus
5054 * an IPv6 header, all options, and a maximal TCP or UDP header.
5056 #define MAX_IPV6_HDR_LEN 256
5058 #define OPT_HDR(type, skb, off) \
5059 (type *)(skb_network_header(skb) + (off))
5061 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
5074 off = sizeof(struct ipv6hdr);
5076 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
5080 nexthdr = ipv6_hdr(skb)->nexthdr;
5082 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
5083 while (off <= len && !done) {
5085 case IPPROTO_DSTOPTS:
5086 case IPPROTO_HOPOPTS:
5087 case IPPROTO_ROUTING: {
5088 struct ipv6_opt_hdr *hp;
5090 err = skb_maybe_pull_tail(skb,
5092 sizeof(struct ipv6_opt_hdr),
5097 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
5098 nexthdr = hp->nexthdr;
5099 off += ipv6_optlen(hp);
5103 struct ip_auth_hdr *hp;
5105 err = skb_maybe_pull_tail(skb,
5107 sizeof(struct ip_auth_hdr),
5112 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
5113 nexthdr = hp->nexthdr;
5114 off += ipv6_authlen(hp);
5117 case IPPROTO_FRAGMENT: {
5118 struct frag_hdr *hp;
5120 err = skb_maybe_pull_tail(skb,
5122 sizeof(struct frag_hdr),
5127 hp = OPT_HDR(struct frag_hdr, skb, off);
5129 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
5132 nexthdr = hp->nexthdr;
5133 off += sizeof(struct frag_hdr);
5144 if (!done || fragment)
5147 csum = skb_checksum_setup_ip(skb, nexthdr, off);
5149 return PTR_ERR(csum);
5152 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5153 &ipv6_hdr(skb)->daddr,
5154 skb->len - off, nexthdr, 0);
5162 * skb_checksum_setup - set up partial checksum offset
5163 * @skb: the skb to set up
5164 * @recalculate: if true the pseudo-header checksum will be recalculated
5166 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
5170 switch (skb->protocol) {
5171 case htons(ETH_P_IP):
5172 err = skb_checksum_setup_ipv4(skb, recalculate);
5175 case htons(ETH_P_IPV6):
5176 err = skb_checksum_setup_ipv6(skb, recalculate);
5186 EXPORT_SYMBOL(skb_checksum_setup);
5189 * skb_checksum_maybe_trim - maybe trims the given skb
5190 * @skb: the skb to check
5191 * @transport_len: the data length beyond the network header
5193 * Checks whether the given skb has data beyond the given transport length.
5194 * If so, returns a cloned skb trimmed to this transport length.
5195 * Otherwise returns the provided skb. Returns NULL in error cases
5196 * (e.g. transport_len exceeds skb length or out-of-memory).
5198 * Caller needs to set the skb transport header and free any returned skb if it
5199 * differs from the provided skb.
5201 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5202 unsigned int transport_len)
5204 struct sk_buff *skb_chk;
5205 unsigned int len = skb_transport_offset(skb) + transport_len;
5210 else if (skb->len == len)
5213 skb_chk = skb_clone(skb, GFP_ATOMIC);
5217 ret = pskb_trim_rcsum(skb_chk, len);
5227 * skb_checksum_trimmed - validate checksum of an skb
5228 * @skb: the skb to check
5229 * @transport_len: the data length beyond the network header
5230 * @skb_chkf: checksum function to use
5232 * Applies the given checksum function skb_chkf to the provided skb.
5233 * Returns a checked and maybe trimmed skb. Returns NULL on error.
5235 * If the skb has data beyond the given transport length, then a
5236 * trimmed & cloned skb is checked and returned.
5238 * Caller needs to set the skb transport header and free any returned skb if it
5239 * differs from the provided skb.
5241 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5242 unsigned int transport_len,
5243 __sum16(*skb_chkf)(struct sk_buff *skb))
5245 struct sk_buff *skb_chk;
5246 unsigned int offset = skb_transport_offset(skb);
5249 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5253 if (!pskb_may_pull(skb_chk, offset))
5256 skb_pull_rcsum(skb_chk, offset);
5257 ret = skb_chkf(skb_chk);
5258 skb_push_rcsum(skb_chk, offset);
5266 if (skb_chk && skb_chk != skb)
5272 EXPORT_SYMBOL(skb_checksum_trimmed);
5274 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5276 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5279 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5281 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5284 skb_release_head_state(skb);
5285 kmem_cache_free(skbuff_head_cache, skb);
5290 EXPORT_SYMBOL(kfree_skb_partial);
5293 * skb_try_coalesce - try to merge skb to prior one
5295 * @from: buffer to add
5296 * @fragstolen: pointer to boolean
5297 * @delta_truesize: how much more was allocated than was requested
5299 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5300 bool *fragstolen, int *delta_truesize)
5302 struct skb_shared_info *to_shinfo, *from_shinfo;
5303 int i, delta, len = from->len;
5305 *fragstolen = false;
5310 /* The page pool signature of struct page will eventually figure out
5311 * which pages can be recycled or not but for now let's prohibit slab
5312 * allocated and page_pool allocated SKBs from being coalesced.
5314 if (to->pp_recycle != from->pp_recycle)
5317 if (len <= skb_tailroom(to)) {
5319 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5320 *delta_truesize = 0;
5324 to_shinfo = skb_shinfo(to);
5325 from_shinfo = skb_shinfo(from);
5326 if (to_shinfo->frag_list || from_shinfo->frag_list)
5328 if (skb_zcopy(to) || skb_zcopy(from))
5331 if (skb_headlen(from) != 0) {
5333 unsigned int offset;
5335 if (to_shinfo->nr_frags +
5336 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5339 if (skb_head_is_locked(from))
5342 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5344 page = virt_to_head_page(from->head);
5345 offset = from->data - (unsigned char *)page_address(page);
5347 skb_fill_page_desc(to, to_shinfo->nr_frags,
5348 page, offset, skb_headlen(from));
5351 if (to_shinfo->nr_frags +
5352 from_shinfo->nr_frags > MAX_SKB_FRAGS)
5355 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5358 WARN_ON_ONCE(delta < len);
5360 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5362 from_shinfo->nr_frags * sizeof(skb_frag_t));
5363 to_shinfo->nr_frags += from_shinfo->nr_frags;
5365 if (!skb_cloned(from))
5366 from_shinfo->nr_frags = 0;
5368 /* if the skb is not cloned this does nothing
5369 * since we set nr_frags to 0.
5371 for (i = 0; i < from_shinfo->nr_frags; i++)
5372 __skb_frag_ref(&from_shinfo->frags[i]);
5374 to->truesize += delta;
5376 to->data_len += len;
5378 *delta_truesize = delta;
5381 EXPORT_SYMBOL(skb_try_coalesce);
5384 * skb_scrub_packet - scrub an skb
5386 * @skb: buffer to clean
5387 * @xnet: packet is crossing netns
5389 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5390 * into/from a tunnel. Some information have to be cleared during these
5392 * skb_scrub_packet can also be used to clean a skb before injecting it in
5393 * another namespace (@xnet == true). We have to clear all information in the
5394 * skb that could impact namespace isolation.
5396 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5398 skb->pkt_type = PACKET_HOST;
5404 nf_reset_trace(skb);
5406 #ifdef CONFIG_NET_SWITCHDEV
5407 skb->offload_fwd_mark = 0;
5408 skb->offload_l3_fwd_mark = 0;
5418 EXPORT_SYMBOL_GPL(skb_scrub_packet);
5421 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5425 * skb_gso_transport_seglen is used to determine the real size of the
5426 * individual segments, including Layer4 headers (TCP/UDP).
5428 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5430 static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5432 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5433 unsigned int thlen = 0;
5435 if (skb->encapsulation) {
5436 thlen = skb_inner_transport_header(skb) -
5437 skb_transport_header(skb);
5439 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5440 thlen += inner_tcp_hdrlen(skb);
5441 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5442 thlen = tcp_hdrlen(skb);
5443 } else if (unlikely(skb_is_gso_sctp(skb))) {
5444 thlen = sizeof(struct sctphdr);
5445 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5446 thlen = sizeof(struct udphdr);
5448 /* UFO sets gso_size to the size of the fragmentation
5449 * payload, i.e. the size of the L4 (UDP) header is already
5452 return thlen + shinfo->gso_size;
5456 * skb_gso_network_seglen - Return length of individual segments of a gso packet
5460 * skb_gso_network_seglen is used to determine the real size of the
5461 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5463 * The MAC/L2 header is not accounted for.
5465 static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5467 unsigned int hdr_len = skb_transport_header(skb) -
5468 skb_network_header(skb);
5470 return hdr_len + skb_gso_transport_seglen(skb);
5474 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5478 * skb_gso_mac_seglen is used to determine the real size of the
5479 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5480 * headers (TCP/UDP).
5482 static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5484 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5486 return hdr_len + skb_gso_transport_seglen(skb);
5490 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5492 * There are a couple of instances where we have a GSO skb, and we
5493 * want to determine what size it would be after it is segmented.
5495 * We might want to check:
5496 * - L3+L4+payload size (e.g. IP forwarding)
5497 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5499 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5503 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5504 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5506 * @max_len: The maximum permissible length.
5508 * Returns true if the segmented length <= max length.
5510 static inline bool skb_gso_size_check(const struct sk_buff *skb,
5511 unsigned int seg_len,
5512 unsigned int max_len) {
5513 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5514 const struct sk_buff *iter;
5516 if (shinfo->gso_size != GSO_BY_FRAGS)
5517 return seg_len <= max_len;
5519 /* Undo this so we can re-use header sizes */
5520 seg_len -= GSO_BY_FRAGS;
5522 skb_walk_frags(skb, iter) {
5523 if (seg_len + skb_headlen(iter) > max_len)
5531 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5534 * @mtu: MTU to validate against
5536 * skb_gso_validate_network_len validates if a given skb will fit a
5537 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5540 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5542 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5544 EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5547 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5550 * @len: length to validate against
5552 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5553 * length once split, including L2, L3 and L4 headers and the payload.
5555 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5557 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5559 EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5561 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5563 int mac_len, meta_len;
5566 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5571 mac_len = skb->data - skb_mac_header(skb);
5572 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5573 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5574 mac_len - VLAN_HLEN - ETH_TLEN);
5577 meta_len = skb_metadata_len(skb);
5579 meta = skb_metadata_end(skb) - meta_len;
5580 memmove(meta + VLAN_HLEN, meta, meta_len);
5583 skb->mac_header += VLAN_HLEN;
5587 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5589 struct vlan_hdr *vhdr;
5592 if (unlikely(skb_vlan_tag_present(skb))) {
5593 /* vlan_tci is already set-up so leave this for another time */
5597 skb = skb_share_check(skb, GFP_ATOMIC);
5600 /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
5601 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
5604 vhdr = (struct vlan_hdr *)skb->data;
5605 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5606 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5608 skb_pull_rcsum(skb, VLAN_HLEN);
5609 vlan_set_encap_proto(skb, vhdr);
5611 skb = skb_reorder_vlan_header(skb);
5615 skb_reset_network_header(skb);
5616 if (!skb_transport_header_was_set(skb))
5617 skb_reset_transport_header(skb);
5618 skb_reset_mac_len(skb);
5626 EXPORT_SYMBOL(skb_vlan_untag);
5628 int skb_ensure_writable(struct sk_buff *skb, int write_len)
5630 if (!pskb_may_pull(skb, write_len))
5633 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5636 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5638 EXPORT_SYMBOL(skb_ensure_writable);
5640 /* remove VLAN header from packet and update csum accordingly.
5641 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5643 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5645 struct vlan_hdr *vhdr;
5646 int offset = skb->data - skb_mac_header(skb);
5649 if (WARN_ONCE(offset,
5650 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5655 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5659 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5661 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5662 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5664 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5665 __skb_pull(skb, VLAN_HLEN);
5667 vlan_set_encap_proto(skb, vhdr);
5668 skb->mac_header += VLAN_HLEN;
5670 if (skb_network_offset(skb) < ETH_HLEN)
5671 skb_set_network_header(skb, ETH_HLEN);
5673 skb_reset_mac_len(skb);
5677 EXPORT_SYMBOL(__skb_vlan_pop);
5679 /* Pop a vlan tag either from hwaccel or from payload.
5680 * Expects skb->data at mac header.
5682 int skb_vlan_pop(struct sk_buff *skb)
5688 if (likely(skb_vlan_tag_present(skb))) {
5689 __vlan_hwaccel_clear_tag(skb);
5691 if (unlikely(!eth_type_vlan(skb->protocol)))
5694 err = __skb_vlan_pop(skb, &vlan_tci);
5698 /* move next vlan tag to hw accel tag */
5699 if (likely(!eth_type_vlan(skb->protocol)))
5702 vlan_proto = skb->protocol;
5703 err = __skb_vlan_pop(skb, &vlan_tci);
5707 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5710 EXPORT_SYMBOL(skb_vlan_pop);
5712 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5713 * Expects skb->data at mac header.
5715 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5717 if (skb_vlan_tag_present(skb)) {
5718 int offset = skb->data - skb_mac_header(skb);
5721 if (WARN_ONCE(offset,
5722 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5727 err = __vlan_insert_tag(skb, skb->vlan_proto,
5728 skb_vlan_tag_get(skb));
5732 skb->protocol = skb->vlan_proto;
5733 skb->mac_len += VLAN_HLEN;
5735 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5737 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5740 EXPORT_SYMBOL(skb_vlan_push);
5743 * skb_eth_pop() - Drop the Ethernet header at the head of a packet
5745 * @skb: Socket buffer to modify
5747 * Drop the Ethernet header of @skb.
5749 * Expects that skb->data points to the mac header and that no VLAN tags are
5752 * Returns 0 on success, -errno otherwise.
5754 int skb_eth_pop(struct sk_buff *skb)
5756 if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) ||
5757 skb_network_offset(skb) < ETH_HLEN)
5760 skb_pull_rcsum(skb, ETH_HLEN);
5761 skb_reset_mac_header(skb);
5762 skb_reset_mac_len(skb);
5766 EXPORT_SYMBOL(skb_eth_pop);
5769 * skb_eth_push() - Add a new Ethernet header at the head of a packet
5771 * @skb: Socket buffer to modify
5772 * @dst: Destination MAC address of the new header
5773 * @src: Source MAC address of the new header
5775 * Prepend @skb with a new Ethernet header.
5777 * Expects that skb->data points to the mac header, which must be empty.
5779 * Returns 0 on success, -errno otherwise.
5781 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
5782 const unsigned char *src)
5787 if (skb_network_offset(skb) || skb_vlan_tag_present(skb))
5790 err = skb_cow_head(skb, sizeof(*eth));
5794 skb_push(skb, sizeof(*eth));
5795 skb_reset_mac_header(skb);
5796 skb_reset_mac_len(skb);
5799 ether_addr_copy(eth->h_dest, dst);
5800 ether_addr_copy(eth->h_source, src);
5801 eth->h_proto = skb->protocol;
5803 skb_postpush_rcsum(skb, eth, sizeof(*eth));
5807 EXPORT_SYMBOL(skb_eth_push);
5809 /* Update the ethertype of hdr and the skb csum value if required. */
5810 static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5813 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5814 __be16 diff[] = { ~hdr->h_proto, ethertype };
5816 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5819 hdr->h_proto = ethertype;
5823 * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
5827 * @mpls_lse: MPLS label stack entry to push
5828 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
5829 * @mac_len: length of the MAC header
5830 * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
5833 * Expects skb->data at mac header.
5835 * Returns 0 on success, -errno otherwise.
5837 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
5838 int mac_len, bool ethernet)
5840 struct mpls_shim_hdr *lse;
5843 if (unlikely(!eth_p_mpls(mpls_proto)))
5846 /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
5847 if (skb->encapsulation)
5850 err = skb_cow_head(skb, MPLS_HLEN);
5854 if (!skb->inner_protocol) {
5855 skb_set_inner_network_header(skb, skb_network_offset(skb));
5856 skb_set_inner_protocol(skb, skb->protocol);
5859 skb_push(skb, MPLS_HLEN);
5860 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
5862 skb_reset_mac_header(skb);
5863 skb_set_network_header(skb, mac_len);
5864 skb_reset_mac_len(skb);
5866 lse = mpls_hdr(skb);
5867 lse->label_stack_entry = mpls_lse;
5868 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
5870 if (ethernet && mac_len >= ETH_HLEN)
5871 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
5872 skb->protocol = mpls_proto;
5876 EXPORT_SYMBOL_GPL(skb_mpls_push);
5879 * skb_mpls_pop() - pop the outermost MPLS header
5882 * @next_proto: ethertype of header after popped MPLS header
5883 * @mac_len: length of the MAC header
5884 * @ethernet: flag to indicate if the packet is ethernet
5886 * Expects skb->data at mac header.
5888 * Returns 0 on success, -errno otherwise.
5890 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
5895 if (unlikely(!eth_p_mpls(skb->protocol)))
5898 err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
5902 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
5903 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
5906 __skb_pull(skb, MPLS_HLEN);
5907 skb_reset_mac_header(skb);
5908 skb_set_network_header(skb, mac_len);
5910 if (ethernet && mac_len >= ETH_HLEN) {
5913 /* use mpls_hdr() to get ethertype to account for VLANs. */
5914 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
5915 skb_mod_eth_type(skb, hdr, next_proto);
5917 skb->protocol = next_proto;
5921 EXPORT_SYMBOL_GPL(skb_mpls_pop);
5924 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
5927 * @mpls_lse: new MPLS label stack entry to update to
5929 * Expects skb->data at mac header.
5931 * Returns 0 on success, -errno otherwise.
5933 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
5937 if (unlikely(!eth_p_mpls(skb->protocol)))
5940 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
5944 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5945 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
5947 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5950 mpls_hdr(skb)->label_stack_entry = mpls_lse;
5954 EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
5957 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
5961 * Expects skb->data at mac header.
5963 * Returns 0 on success, -errno otherwise.
5965 int skb_mpls_dec_ttl(struct sk_buff *skb)
5970 if (unlikely(!eth_p_mpls(skb->protocol)))
5973 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
5976 lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
5977 ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
5981 lse &= ~MPLS_LS_TTL_MASK;
5982 lse |= ttl << MPLS_LS_TTL_SHIFT;
5984 return skb_mpls_update_lse(skb, cpu_to_be32(lse));
5986 EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
5989 * alloc_skb_with_frags - allocate skb with page frags
5991 * @header_len: size of linear part
5992 * @data_len: needed length in frags
5993 * @max_page_order: max page order desired.
5994 * @errcode: pointer to error code if any
5995 * @gfp_mask: allocation mask
5997 * This can be used to allocate a paged skb, given a maximal order for frags.
5999 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
6000 unsigned long data_len,
6005 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
6006 unsigned long chunk;
6007 struct sk_buff *skb;
6011 *errcode = -EMSGSIZE;
6012 /* Note this test could be relaxed, if we succeed to allocate
6013 * high order pages...
6015 if (npages > MAX_SKB_FRAGS)
6018 *errcode = -ENOBUFS;
6019 skb = alloc_skb(header_len, gfp_mask);
6023 skb->truesize += npages << PAGE_SHIFT;
6025 for (i = 0; npages > 0; i++) {
6026 int order = max_page_order;
6029 if (npages >= 1 << order) {
6030 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
6036 /* Do not retry other high order allocations */
6042 page = alloc_page(gfp_mask);
6046 chunk = min_t(unsigned long, data_len,
6047 PAGE_SIZE << order);
6048 skb_fill_page_desc(skb, i, page, 0, chunk);
6050 npages -= 1 << order;
6058 EXPORT_SYMBOL(alloc_skb_with_frags);
6060 /* carve out the first off bytes from skb when off < headlen */
6061 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
6062 const int headlen, gfp_t gfp_mask)
6065 int size = skb_end_offset(skb);
6066 int new_hlen = headlen - off;
6069 size = SKB_DATA_ALIGN(size);
6071 if (skb_pfmemalloc(skb))
6072 gfp_mask |= __GFP_MEMALLOC;
6073 data = kmalloc_reserve(size +
6074 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
6075 gfp_mask, NUMA_NO_NODE, NULL);
6079 size = SKB_WITH_OVERHEAD(ksize(data));
6081 /* Copy real data, and all frags */
6082 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
6085 memcpy((struct skb_shared_info *)(data + size),
6087 offsetof(struct skb_shared_info,
6088 frags[skb_shinfo(skb)->nr_frags]));
6089 if (skb_cloned(skb)) {
6090 /* drop the old head gracefully */
6091 if (skb_orphan_frags(skb, gfp_mask)) {
6095 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
6096 skb_frag_ref(skb, i);
6097 if (skb_has_frag_list(skb))
6098 skb_clone_fraglist(skb);
6099 skb_release_data(skb);
6101 /* we can reuse existing recount- all we did was
6110 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6113 skb->end = skb->head + size;
6115 skb_set_tail_pointer(skb, skb_headlen(skb));
6116 skb_headers_offset_update(skb, 0);
6120 atomic_set(&skb_shinfo(skb)->dataref, 1);
6125 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
6127 /* carve out the first eat bytes from skb's frag_list. May recurse into
6130 static int pskb_carve_frag_list(struct sk_buff *skb,
6131 struct skb_shared_info *shinfo, int eat,
6134 struct sk_buff *list = shinfo->frag_list;
6135 struct sk_buff *clone = NULL;
6136 struct sk_buff *insp = NULL;
6140 pr_err("Not enough bytes to eat. Want %d\n", eat);
6143 if (list->len <= eat) {
6144 /* Eaten as whole. */
6149 /* Eaten partially. */
6150 if (skb_shared(list)) {
6151 clone = skb_clone(list, gfp_mask);
6157 /* This may be pulled without problems. */
6160 if (pskb_carve(list, eat, gfp_mask) < 0) {
6168 /* Free pulled out fragments. */
6169 while ((list = shinfo->frag_list) != insp) {
6170 shinfo->frag_list = list->next;
6173 /* And insert new clone at head. */
6176 shinfo->frag_list = clone;
6181 /* carve off first len bytes from skb. Split line (off) is in the
6182 * non-linear part of skb
6184 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
6185 int pos, gfp_t gfp_mask)
6188 int size = skb_end_offset(skb);
6190 const int nfrags = skb_shinfo(skb)->nr_frags;
6191 struct skb_shared_info *shinfo;
6193 size = SKB_DATA_ALIGN(size);
6195 if (skb_pfmemalloc(skb))
6196 gfp_mask |= __GFP_MEMALLOC;
6197 data = kmalloc_reserve(size +
6198 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
6199 gfp_mask, NUMA_NO_NODE, NULL);
6203 size = SKB_WITH_OVERHEAD(ksize(data));
6205 memcpy((struct skb_shared_info *)(data + size),
6206 skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0]));
6207 if (skb_orphan_frags(skb, gfp_mask)) {
6211 shinfo = (struct skb_shared_info *)(data + size);
6212 for (i = 0; i < nfrags; i++) {
6213 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
6215 if (pos + fsize > off) {
6216 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
6220 * We have two variants in this case:
6221 * 1. Move all the frag to the second
6222 * part, if it is possible. F.e.
6223 * this approach is mandatory for TUX,
6224 * where splitting is expensive.
6225 * 2. Split is accurately. We make this.
6227 skb_frag_off_add(&shinfo->frags[0], off - pos);
6228 skb_frag_size_sub(&shinfo->frags[0], off - pos);
6230 skb_frag_ref(skb, i);
6235 shinfo->nr_frags = k;
6236 if (skb_has_frag_list(skb))
6237 skb_clone_fraglist(skb);
6239 /* split line is in frag list */
6240 if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
6241 /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
6242 if (skb_has_frag_list(skb))
6243 kfree_skb_list(skb_shinfo(skb)->frag_list);
6247 skb_release_data(skb);
6252 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6255 skb->end = skb->head + size;
6257 skb_reset_tail_pointer(skb);
6258 skb_headers_offset_update(skb, 0);
6263 skb->data_len = skb->len;
6264 atomic_set(&skb_shinfo(skb)->dataref, 1);
6268 /* remove len bytes from the beginning of the skb */
6269 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6271 int headlen = skb_headlen(skb);
6274 return pskb_carve_inside_header(skb, len, headlen, gfp);
6276 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6279 /* Extract to_copy bytes starting at off from skb, and return this in
6282 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6283 int to_copy, gfp_t gfp)
6285 struct sk_buff *clone = skb_clone(skb, gfp);
6290 if (pskb_carve(clone, off, gfp) < 0 ||
6291 pskb_trim(clone, to_copy)) {
6297 EXPORT_SYMBOL(pskb_extract);
6300 * skb_condense - try to get rid of fragments/frag_list if possible
6303 * Can be used to save memory before skb is added to a busy queue.
6304 * If packet has bytes in frags and enough tail room in skb->head,
6305 * pull all of them, so that we can free the frags right now and adjust
6308 * We do not reallocate skb->head thus can not fail.
6309 * Caller must re-evaluate skb->truesize if needed.
6311 void skb_condense(struct sk_buff *skb)
6313 if (skb->data_len) {
6314 if (skb->data_len > skb->end - skb->tail ||
6318 /* Nice, we can free page frag(s) right now */
6319 __pskb_pull_tail(skb, skb->data_len);
6321 /* At this point, skb->truesize might be over estimated,
6322 * because skb had a fragment, and fragments do not tell
6324 * When we pulled its content into skb->head, fragment
6325 * was freed, but __pskb_pull_tail() could not possibly
6326 * adjust skb->truesize, not knowing the frag truesize.
6328 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6331 #ifdef CONFIG_SKB_EXTENSIONS
6332 static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6334 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6338 * __skb_ext_alloc - allocate a new skb extensions storage
6340 * @flags: See kmalloc().
6342 * Returns the newly allocated pointer. The pointer can later attached to a
6343 * skb via __skb_ext_set().
6344 * Note: caller must handle the skb_ext as an opaque data.
6346 struct skb_ext *__skb_ext_alloc(gfp_t flags)
6348 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6351 memset(new->offset, 0, sizeof(new->offset));
6352 refcount_set(&new->refcnt, 1);
6358 static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6359 unsigned int old_active)
6361 struct skb_ext *new;
6363 if (refcount_read(&old->refcnt) == 1)
6366 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6370 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6371 refcount_set(&new->refcnt, 1);
6374 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6375 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6378 for (i = 0; i < sp->len; i++)
6379 xfrm_state_hold(sp->xvec[i]);
6387 * __skb_ext_set - attach the specified extension storage to this skb
6390 * @ext: extension storage previously allocated via __skb_ext_alloc()
6392 * Existing extensions, if any, are cleared.
6394 * Returns the pointer to the extension.
6396 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6397 struct skb_ext *ext)
6399 unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6402 newlen = newoff + skb_ext_type_len[id];
6403 ext->chunks = newlen;
6404 ext->offset[id] = newoff;
6405 skb->extensions = ext;
6406 skb->active_extensions = 1 << id;
6407 return skb_ext_get_ptr(ext, id);
6411 * skb_ext_add - allocate space for given extension, COW if needed
6413 * @id: extension to allocate space for
6415 * Allocates enough space for the given extension.
6416 * If the extension is already present, a pointer to that extension
6419 * If the skb was cloned, COW applies and the returned memory can be
6420 * modified without changing the extension space of clones buffers.
6422 * Returns pointer to the extension or NULL on allocation failure.
6424 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6426 struct skb_ext *new, *old = NULL;
6427 unsigned int newlen, newoff;
6429 if (skb->active_extensions) {
6430 old = skb->extensions;
6432 new = skb_ext_maybe_cow(old, skb->active_extensions);
6436 if (__skb_ext_exist(new, id))
6439 newoff = new->chunks;
6441 newoff = SKB_EXT_CHUNKSIZEOF(*new);
6443 new = __skb_ext_alloc(GFP_ATOMIC);
6448 newlen = newoff + skb_ext_type_len[id];
6449 new->chunks = newlen;
6450 new->offset[id] = newoff;
6452 skb->extensions = new;
6453 skb->active_extensions |= 1 << id;
6454 return skb_ext_get_ptr(new, id);
6456 EXPORT_SYMBOL(skb_ext_add);
6459 static void skb_ext_put_sp(struct sec_path *sp)
6463 for (i = 0; i < sp->len; i++)
6464 xfrm_state_put(sp->xvec[i]);
6468 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6470 struct skb_ext *ext = skb->extensions;
6472 skb->active_extensions &= ~(1 << id);
6473 if (skb->active_extensions == 0) {
6474 skb->extensions = NULL;
6477 } else if (id == SKB_EXT_SEC_PATH &&
6478 refcount_read(&ext->refcnt) == 1) {
6479 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6486 EXPORT_SYMBOL(__skb_ext_del);
6488 void __skb_ext_put(struct skb_ext *ext)
6490 /* If this is last clone, nothing can increment
6491 * it after check passes. Avoids one atomic op.
6493 if (refcount_read(&ext->refcnt) == 1)
6496 if (!refcount_dec_and_test(&ext->refcnt))
6500 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6501 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6504 kmem_cache_free(skbuff_ext_cache, ext);
6506 EXPORT_SYMBOL(__skb_ext_put);
6507 #endif /* CONFIG_SKB_EXTENSIONS */