1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Routines having to do with the 'struct sk_buff' memory handlers.
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Florian La Roche <rzsfl@rz.uni-sb.de>
9 * Alan Cox : Fixed the worst of the load
11 * Dave Platt : Interrupt stacking fix.
12 * Richard Kooijman : Timestamp fixes.
13 * Alan Cox : Changed buffer format.
14 * Alan Cox : destructor hook for AF_UNIX etc.
15 * Linus Torvalds : Better skb_clone.
16 * Alan Cox : Added skb_copy.
17 * Alan Cox : Added all the changed routines Linus
18 * only put in the headers
19 * Ray VanTassle : Fixed --skb->lock in free
20 * Alan Cox : skb_copy copy arp field
21 * Andi Kleen : slabified it.
22 * Robert Olsson : Removed skb_head_pool
25 * The __skb_ routines should be called with interrupts
26 * disabled, or you better be *real* sure that the operation is atomic
27 * with respect to whatever list is being frobbed (e.g. via lock_sock()
28 * or via disabling bottom half handlers, etc).
32 * The functions in this file will not compile correctly with gcc 2.4.x
35 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/kernel.h>
41 #include <linux/interrupt.h>
43 #include <linux/inet.h>
44 #include <linux/slab.h>
45 #include <linux/tcp.h>
46 #include <linux/udp.h>
47 #include <linux/sctp.h>
48 #include <linux/netdevice.h>
49 #ifdef CONFIG_NET_CLS_ACT
50 #include <net/pkt_sched.h>
52 #include <linux/string.h>
53 #include <linux/skbuff.h>
54 #include <linux/splice.h>
55 #include <linux/cache.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/init.h>
58 #include <linux/scatterlist.h>
59 #include <linux/errqueue.h>
60 #include <linux/prefetch.h>
61 #include <linux/if_vlan.h>
62 #include <linux/mpls.h>
63 #include <linux/kcov.h>
65 #include <net/protocol.h>
68 #include <net/checksum.h>
69 #include <net/ip6_checksum.h>
72 #include <net/mptcp.h>
74 #include <net/page_pool.h>
76 #include <linux/uaccess.h>
77 #include <trace/events/skb.h>
78 #include <linux/highmem.h>
79 #include <linux/capability.h>
80 #include <linux/user_namespace.h>
81 #include <linux/indirect_call_wrapper.h>
84 #include "sock_destructor.h"
86 struct kmem_cache *skbuff_head_cache __ro_after_init;
87 static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
88 #ifdef CONFIG_SKB_EXTENSIONS
89 static struct kmem_cache *skbuff_ext_cache __ro_after_init;
91 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
92 EXPORT_SYMBOL(sysctl_max_skb_frags);
95 #define FN(reason) [SKB_DROP_REASON_##reason] = #reason,
96 const char * const drop_reasons[] = {
97 DEFINE_DROP_REASON(FN, FN)
99 EXPORT_SYMBOL(drop_reasons);
102 * skb_panic - private function for out-of-line support
106 * @msg: skb_over_panic or skb_under_panic
108 * Out-of-line support for skb_put() and skb_push().
109 * Called via the wrapper skb_over_panic() or skb_under_panic().
110 * Keep out of line to prevent kernel bloat.
111 * __builtin_return_address is not used because it is not always reliable.
113 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
116 pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n",
117 msg, addr, skb->len, sz, skb->head, skb->data,
118 (unsigned long)skb->tail, (unsigned long)skb->end,
119 skb->dev ? skb->dev->name : "<NULL>");
123 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
125 skb_panic(skb, sz, addr, __func__);
128 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
130 skb_panic(skb, sz, addr, __func__);
133 #define NAPI_SKB_CACHE_SIZE 64
134 #define NAPI_SKB_CACHE_BULK 16
135 #define NAPI_SKB_CACHE_HALF (NAPI_SKB_CACHE_SIZE / 2)
137 #if PAGE_SIZE == SZ_4K
139 #define NAPI_HAS_SMALL_PAGE_FRAG 1
140 #define NAPI_SMALL_PAGE_PFMEMALLOC(nc) ((nc).pfmemalloc)
142 /* specialized page frag allocator using a single order 0 page
143 * and slicing it into 1K sized fragment. Constrained to systems
144 * with a very limited amount of 1K fragments fitting a single
145 * page - to avoid excessive truesize underestimation
148 struct page_frag_1k {
154 static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp)
159 offset = nc->offset - SZ_1K;
160 if (likely(offset >= 0))
163 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
167 nc->va = page_address(page);
168 nc->pfmemalloc = page_is_pfmemalloc(page);
169 offset = PAGE_SIZE - SZ_1K;
170 page_ref_add(page, offset / SZ_1K);
174 return nc->va + offset;
178 /* the small page is actually unused in this build; add dummy helpers
179 * to please the compiler and avoid later preprocessor's conditionals
181 #define NAPI_HAS_SMALL_PAGE_FRAG 0
182 #define NAPI_SMALL_PAGE_PFMEMALLOC(nc) false
184 struct page_frag_1k {
187 static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp_mask)
194 struct napi_alloc_cache {
195 struct page_frag_cache page;
196 struct page_frag_1k page_small;
197 unsigned int skb_count;
198 void *skb_cache[NAPI_SKB_CACHE_SIZE];
201 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
202 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
204 /* Double check that napi_get_frags() allocates skbs with
205 * skb->head being backed by slab, not a page fragment.
206 * This is to make sure bug fixed in 3226b158e67c
207 * ("net: avoid 32 x truesize under-estimation for tiny skbs")
208 * does not accidentally come back.
210 void napi_get_frags_check(struct napi_struct *napi)
215 skb = napi_get_frags(napi);
216 WARN_ON_ONCE(!NAPI_HAS_SMALL_PAGE_FRAG && skb && skb->head_frag);
217 napi_free_frags(napi);
221 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
223 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
225 fragsz = SKB_DATA_ALIGN(fragsz);
227 return page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask);
229 EXPORT_SYMBOL(__napi_alloc_frag_align);
231 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
235 fragsz = SKB_DATA_ALIGN(fragsz);
236 if (in_hardirq() || irqs_disabled()) {
237 struct page_frag_cache *nc = this_cpu_ptr(&netdev_alloc_cache);
239 data = page_frag_alloc_align(nc, fragsz, GFP_ATOMIC, align_mask);
241 struct napi_alloc_cache *nc;
244 nc = this_cpu_ptr(&napi_alloc_cache);
245 data = page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask);
250 EXPORT_SYMBOL(__netdev_alloc_frag_align);
252 static struct sk_buff *napi_skb_cache_get(void)
254 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
257 if (unlikely(!nc->skb_count)) {
258 nc->skb_count = kmem_cache_alloc_bulk(skbuff_head_cache,
262 if (unlikely(!nc->skb_count))
266 skb = nc->skb_cache[--nc->skb_count];
267 kasan_unpoison_object_data(skbuff_head_cache, skb);
272 /* Caller must provide SKB that is memset cleared */
273 static void __build_skb_around(struct sk_buff *skb, void *data,
274 unsigned int frag_size)
276 struct skb_shared_info *shinfo;
277 unsigned int size = frag_size ? : ksize(data);
279 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
281 /* Assumes caller memset cleared SKB */
282 skb->truesize = SKB_TRUESIZE(size);
283 refcount_set(&skb->users, 1);
286 skb_reset_tail_pointer(skb);
287 skb_set_end_offset(skb, size);
288 skb->mac_header = (typeof(skb->mac_header))~0U;
289 skb->transport_header = (typeof(skb->transport_header))~0U;
290 skb->alloc_cpu = raw_smp_processor_id();
291 /* make sure we initialize shinfo sequentially */
292 shinfo = skb_shinfo(skb);
293 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
294 atomic_set(&shinfo->dataref, 1);
296 skb_set_kcov_handle(skb, kcov_common_handle());
300 * __build_skb - build a network buffer
301 * @data: data buffer provided by caller
302 * @frag_size: size of data, or 0 if head was kmalloced
304 * Allocate a new &sk_buff. Caller provides space holding head and
305 * skb_shared_info. @data must have been allocated by kmalloc() only if
306 * @frag_size is 0, otherwise data should come from the page allocator
308 * The return is the new skb buffer.
309 * On a failure the return is %NULL, and @data is not freed.
311 * Before IO, driver allocates only data buffer where NIC put incoming frame
312 * Driver should add room at head (NET_SKB_PAD) and
313 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
314 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
315 * before giving packet to stack.
316 * RX rings only contains data buffers, not full skbs.
318 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
322 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
326 memset(skb, 0, offsetof(struct sk_buff, tail));
327 __build_skb_around(skb, data, frag_size);
332 /* build_skb() is wrapper over __build_skb(), that specifically
333 * takes care of skb->head and skb->pfmemalloc
334 * This means that if @frag_size is not zero, then @data must be backed
335 * by a page fragment, not kmalloc() or vmalloc()
337 struct sk_buff *build_skb(void *data, unsigned int frag_size)
339 struct sk_buff *skb = __build_skb(data, frag_size);
341 if (skb && frag_size) {
343 if (page_is_pfmemalloc(virt_to_head_page(data)))
348 EXPORT_SYMBOL(build_skb);
351 * build_skb_around - build a network buffer around provided skb
352 * @skb: sk_buff provide by caller, must be memset cleared
353 * @data: data buffer provided by caller
354 * @frag_size: size of data, or 0 if head was kmalloced
356 struct sk_buff *build_skb_around(struct sk_buff *skb,
357 void *data, unsigned int frag_size)
362 __build_skb_around(skb, data, frag_size);
366 if (page_is_pfmemalloc(virt_to_head_page(data)))
371 EXPORT_SYMBOL(build_skb_around);
374 * __napi_build_skb - build a network buffer
375 * @data: data buffer provided by caller
376 * @frag_size: size of data, or 0 if head was kmalloced
378 * Version of __build_skb() that uses NAPI percpu caches to obtain
379 * skbuff_head instead of inplace allocation.
381 * Returns a new &sk_buff on success, %NULL on allocation failure.
383 static struct sk_buff *__napi_build_skb(void *data, unsigned int frag_size)
387 skb = napi_skb_cache_get();
391 memset(skb, 0, offsetof(struct sk_buff, tail));
392 __build_skb_around(skb, data, frag_size);
398 * napi_build_skb - build a network buffer
399 * @data: data buffer provided by caller
400 * @frag_size: size of data, or 0 if head was kmalloced
402 * Version of __napi_build_skb() that takes care of skb->head_frag
403 * and skb->pfmemalloc when the data is a page or page fragment.
405 * Returns a new &sk_buff on success, %NULL on allocation failure.
407 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size)
409 struct sk_buff *skb = __napi_build_skb(data, frag_size);
411 if (likely(skb) && frag_size) {
413 skb_propagate_pfmemalloc(virt_to_head_page(data), skb);
418 EXPORT_SYMBOL(napi_build_skb);
421 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
422 * the caller if emergency pfmemalloc reserves are being used. If it is and
423 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
424 * may be used. Otherwise, the packet data may be discarded until enough
427 static void *kmalloc_reserve(size_t size, gfp_t flags, int node,
431 bool ret_pfmemalloc = false;
434 * Try a regular allocation, when that fails and we're not entitled
435 * to the reserves, fail.
437 obj = kmalloc_node_track_caller(size,
438 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
440 if (obj || !(gfp_pfmemalloc_allowed(flags)))
443 /* Try again but now we are using pfmemalloc reserves */
444 ret_pfmemalloc = true;
445 obj = kmalloc_node_track_caller(size, flags, node);
449 *pfmemalloc = ret_pfmemalloc;
454 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
455 * 'private' fields and also do memory statistics to find all the
461 * __alloc_skb - allocate a network buffer
462 * @size: size to allocate
463 * @gfp_mask: allocation mask
464 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
465 * instead of head cache and allocate a cloned (child) skb.
466 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
467 * allocations in case the data is required for writeback
468 * @node: numa node to allocate memory on
470 * Allocate a new &sk_buff. The returned buffer has no headroom and a
471 * tail room of at least size bytes. The object has a reference count
472 * of one. The return is the buffer. On a failure the return is %NULL.
474 * Buffers may only be allocated from interrupts using a @gfp_mask of
477 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
480 struct kmem_cache *cache;
486 cache = (flags & SKB_ALLOC_FCLONE)
487 ? skbuff_fclone_cache : skbuff_head_cache;
489 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
490 gfp_mask |= __GFP_MEMALLOC;
493 if ((flags & (SKB_ALLOC_FCLONE | SKB_ALLOC_NAPI)) == SKB_ALLOC_NAPI &&
494 likely(node == NUMA_NO_NODE || node == numa_mem_id()))
495 skb = napi_skb_cache_get();
497 skb = kmem_cache_alloc_node(cache, gfp_mask & ~GFP_DMA, node);
502 /* We do our best to align skb_shared_info on a separate cache
503 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
504 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
505 * Both skb->head and skb_shared_info are cache line aligned.
507 size = SKB_DATA_ALIGN(size);
508 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
509 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
512 /* kmalloc(size) might give us more room than requested.
513 * Put skb_shared_info exactly at the end of allocated zone,
514 * to allow max possible filling before reallocation.
517 size = SKB_WITH_OVERHEAD(osize);
518 prefetchw(data + size);
521 * Only clear those fields we need to clear, not those that we will
522 * actually initialise below. Hence, don't put any more fields after
523 * the tail pointer in struct sk_buff!
525 memset(skb, 0, offsetof(struct sk_buff, tail));
526 __build_skb_around(skb, data, osize);
527 skb->pfmemalloc = pfmemalloc;
529 if (flags & SKB_ALLOC_FCLONE) {
530 struct sk_buff_fclones *fclones;
532 fclones = container_of(skb, struct sk_buff_fclones, skb1);
534 skb->fclone = SKB_FCLONE_ORIG;
535 refcount_set(&fclones->fclone_ref, 1);
541 kmem_cache_free(cache, skb);
544 EXPORT_SYMBOL(__alloc_skb);
547 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
548 * @dev: network device to receive on
549 * @len: length to allocate
550 * @gfp_mask: get_free_pages mask, passed to alloc_skb
552 * Allocate a new &sk_buff and assign it a usage count of one. The
553 * buffer has NET_SKB_PAD headroom built in. Users should allocate
554 * the headroom they think they need without accounting for the
555 * built in space. The built in space is used for optimisations.
557 * %NULL is returned if there is no free memory.
559 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
562 struct page_frag_cache *nc;
569 /* If requested length is either too small or too big,
570 * we use kmalloc() for skb->head allocation.
572 if (len <= SKB_WITH_OVERHEAD(1024) ||
573 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
574 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
575 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
581 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
582 len = SKB_DATA_ALIGN(len);
584 if (sk_memalloc_socks())
585 gfp_mask |= __GFP_MEMALLOC;
587 if (in_hardirq() || irqs_disabled()) {
588 nc = this_cpu_ptr(&netdev_alloc_cache);
589 data = page_frag_alloc(nc, len, gfp_mask);
590 pfmemalloc = nc->pfmemalloc;
593 nc = this_cpu_ptr(&napi_alloc_cache.page);
594 data = page_frag_alloc(nc, len, gfp_mask);
595 pfmemalloc = nc->pfmemalloc;
602 skb = __build_skb(data, len);
603 if (unlikely(!skb)) {
613 skb_reserve(skb, NET_SKB_PAD);
619 EXPORT_SYMBOL(__netdev_alloc_skb);
622 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
623 * @napi: napi instance this buffer was allocated for
624 * @len: length to allocate
625 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
627 * Allocate a new sk_buff for use in NAPI receive. This buffer will
628 * attempt to allocate the head from a special reserved region used
629 * only for NAPI Rx allocation. By doing this we can save several
630 * CPU cycles by avoiding having to disable and re-enable IRQs.
632 * %NULL is returned if there is no free memory.
634 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
637 struct napi_alloc_cache *nc;
642 DEBUG_NET_WARN_ON_ONCE(!in_softirq());
643 len += NET_SKB_PAD + NET_IP_ALIGN;
645 /* If requested length is either too small or too big,
646 * we use kmalloc() for skb->head allocation.
647 * When the small frag allocator is available, prefer it over kmalloc
648 * for small fragments
650 if ((!NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) ||
651 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
652 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
653 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI,
660 nc = this_cpu_ptr(&napi_alloc_cache);
662 if (sk_memalloc_socks())
663 gfp_mask |= __GFP_MEMALLOC;
665 if (NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) {
666 /* we are artificially inflating the allocation size, but
667 * that is not as bad as it may look like, as:
668 * - 'len' less than GRO_MAX_HEAD makes little sense
669 * - On most systems, larger 'len' values lead to fragment
670 * size above 512 bytes
671 * - kmalloc would use the kmalloc-1k slab for such values
672 * - Builds with smaller GRO_MAX_HEAD will very likely do
673 * little networking, as that implies no WiFi and no
674 * tunnels support, and 32 bits arches.
678 data = page_frag_alloc_1k(&nc->page_small, gfp_mask);
679 pfmemalloc = NAPI_SMALL_PAGE_PFMEMALLOC(nc->page_small);
681 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
682 len = SKB_DATA_ALIGN(len);
684 data = page_frag_alloc(&nc->page, len, gfp_mask);
685 pfmemalloc = nc->page.pfmemalloc;
691 skb = __napi_build_skb(data, len);
692 if (unlikely(!skb)) {
702 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
703 skb->dev = napi->dev;
708 EXPORT_SYMBOL(__napi_alloc_skb);
710 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
711 int size, unsigned int truesize)
713 skb_fill_page_desc(skb, i, page, off, size);
715 skb->data_len += size;
716 skb->truesize += truesize;
718 EXPORT_SYMBOL(skb_add_rx_frag);
720 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
721 unsigned int truesize)
723 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
725 skb_frag_size_add(frag, size);
727 skb->data_len += size;
728 skb->truesize += truesize;
730 EXPORT_SYMBOL(skb_coalesce_rx_frag);
732 static void skb_drop_list(struct sk_buff **listp)
734 kfree_skb_list(*listp);
738 static inline void skb_drop_fraglist(struct sk_buff *skb)
740 skb_drop_list(&skb_shinfo(skb)->frag_list);
743 static void skb_clone_fraglist(struct sk_buff *skb)
745 struct sk_buff *list;
747 skb_walk_frags(skb, list)
751 static void skb_free_head(struct sk_buff *skb)
753 unsigned char *head = skb->head;
755 if (skb->head_frag) {
756 if (skb_pp_recycle(skb, head))
764 static void skb_release_data(struct sk_buff *skb)
766 struct skb_shared_info *shinfo = skb_shinfo(skb);
770 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
774 if (skb_zcopy(skb)) {
775 bool skip_unref = shinfo->flags & SKBFL_MANAGED_FRAG_REFS;
777 skb_zcopy_clear(skb, true);
782 for (i = 0; i < shinfo->nr_frags; i++)
783 __skb_frag_unref(&shinfo->frags[i], skb->pp_recycle);
786 if (shinfo->frag_list)
787 kfree_skb_list(shinfo->frag_list);
791 /* When we clone an SKB we copy the reycling bit. The pp_recycle
792 * bit is only set on the head though, so in order to avoid races
793 * while trying to recycle fragments on __skb_frag_unref() we need
794 * to make one SKB responsible for triggering the recycle path.
795 * So disable the recycling bit if an SKB is cloned and we have
796 * additional references to the fragmented part of the SKB.
797 * Eventually the last SKB will have the recycling bit set and it's
798 * dataref set to 0, which will trigger the recycling
804 * Free an skbuff by memory without cleaning the state.
806 static void kfree_skbmem(struct sk_buff *skb)
808 struct sk_buff_fclones *fclones;
810 switch (skb->fclone) {
811 case SKB_FCLONE_UNAVAILABLE:
812 kmem_cache_free(skbuff_head_cache, skb);
815 case SKB_FCLONE_ORIG:
816 fclones = container_of(skb, struct sk_buff_fclones, skb1);
818 /* We usually free the clone (TX completion) before original skb
819 * This test would have no chance to be true for the clone,
820 * while here, branch prediction will be good.
822 if (refcount_read(&fclones->fclone_ref) == 1)
826 default: /* SKB_FCLONE_CLONE */
827 fclones = container_of(skb, struct sk_buff_fclones, skb2);
830 if (!refcount_dec_and_test(&fclones->fclone_ref))
833 kmem_cache_free(skbuff_fclone_cache, fclones);
836 void skb_release_head_state(struct sk_buff *skb)
839 if (skb->destructor) {
840 DEBUG_NET_WARN_ON_ONCE(in_hardirq());
841 skb->destructor(skb);
843 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
844 nf_conntrack_put(skb_nfct(skb));
849 /* Free everything but the sk_buff shell. */
850 static void skb_release_all(struct sk_buff *skb)
852 skb_release_head_state(skb);
853 if (likely(skb->head))
854 skb_release_data(skb);
858 * __kfree_skb - private function
861 * Free an sk_buff. Release anything attached to the buffer.
862 * Clean the state. This is an internal helper function. Users should
863 * always call kfree_skb
866 void __kfree_skb(struct sk_buff *skb)
868 skb_release_all(skb);
871 EXPORT_SYMBOL(__kfree_skb);
874 * kfree_skb_reason - free an sk_buff with special reason
875 * @skb: buffer to free
876 * @reason: reason why this skb is dropped
878 * Drop a reference to the buffer and free it if the usage count has
879 * hit zero. Meanwhile, pass the drop reason to 'kfree_skb'
883 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason)
885 if (unlikely(!skb_unref(skb)))
888 DEBUG_NET_WARN_ON_ONCE(reason <= 0 || reason >= SKB_DROP_REASON_MAX);
890 trace_kfree_skb(skb, __builtin_return_address(0), reason);
893 EXPORT_SYMBOL(kfree_skb_reason);
895 void kfree_skb_list_reason(struct sk_buff *segs,
896 enum skb_drop_reason reason)
899 struct sk_buff *next = segs->next;
901 kfree_skb_reason(segs, reason);
905 EXPORT_SYMBOL(kfree_skb_list_reason);
907 /* Dump skb information and contents.
909 * Must only be called from net_ratelimit()-ed paths.
911 * Dumps whole packets if full_pkt, only headers otherwise.
913 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
915 struct skb_shared_info *sh = skb_shinfo(skb);
916 struct net_device *dev = skb->dev;
917 struct sock *sk = skb->sk;
918 struct sk_buff *list_skb;
919 bool has_mac, has_trans;
920 int headroom, tailroom;
926 len = min_t(int, skb->len, MAX_HEADER + 128);
928 headroom = skb_headroom(skb);
929 tailroom = skb_tailroom(skb);
931 has_mac = skb_mac_header_was_set(skb);
932 has_trans = skb_transport_header_was_set(skb);
934 printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
935 "mac=(%d,%d) net=(%d,%d) trans=%d\n"
936 "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
937 "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
938 "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
939 level, skb->len, headroom, skb_headlen(skb), tailroom,
940 has_mac ? skb->mac_header : -1,
941 has_mac ? skb_mac_header_len(skb) : -1,
943 has_trans ? skb_network_header_len(skb) : -1,
944 has_trans ? skb->transport_header : -1,
945 sh->tx_flags, sh->nr_frags,
946 sh->gso_size, sh->gso_type, sh->gso_segs,
947 skb->csum, skb->ip_summed, skb->csum_complete_sw,
948 skb->csum_valid, skb->csum_level,
949 skb->hash, skb->sw_hash, skb->l4_hash,
950 ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
953 printk("%sdev name=%s feat=%pNF\n",
954 level, dev->name, &dev->features);
956 printk("%ssk family=%hu type=%u proto=%u\n",
957 level, sk->sk_family, sk->sk_type, sk->sk_protocol);
959 if (full_pkt && headroom)
960 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
961 16, 1, skb->head, headroom, false);
963 seg_len = min_t(int, skb_headlen(skb), len);
965 print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET,
966 16, 1, skb->data, seg_len, false);
969 if (full_pkt && tailroom)
970 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
971 16, 1, skb_tail_pointer(skb), tailroom, false);
973 for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
974 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
975 u32 p_off, p_len, copied;
979 skb_frag_foreach_page(frag, skb_frag_off(frag),
980 skb_frag_size(frag), p, p_off, p_len,
982 seg_len = min_t(int, p_len, len);
983 vaddr = kmap_atomic(p);
984 print_hex_dump(level, "skb frag: ",
986 16, 1, vaddr + p_off, seg_len, false);
987 kunmap_atomic(vaddr);
994 if (full_pkt && skb_has_frag_list(skb)) {
995 printk("skb fraglist:\n");
996 skb_walk_frags(skb, list_skb)
997 skb_dump(level, list_skb, true);
1000 EXPORT_SYMBOL(skb_dump);
1003 * skb_tx_error - report an sk_buff xmit error
1004 * @skb: buffer that triggered an error
1006 * Report xmit error if a device callback is tracking this skb.
1007 * skb must be freed afterwards.
1009 void skb_tx_error(struct sk_buff *skb)
1012 skb_zcopy_downgrade_managed(skb);
1013 skb_zcopy_clear(skb, true);
1016 EXPORT_SYMBOL(skb_tx_error);
1018 #ifdef CONFIG_TRACEPOINTS
1020 * consume_skb - free an skbuff
1021 * @skb: buffer to free
1023 * Drop a ref to the buffer and free it if the usage count has hit zero
1024 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
1025 * is being dropped after a failure and notes that
1027 void consume_skb(struct sk_buff *skb)
1029 if (!skb_unref(skb))
1032 trace_consume_skb(skb);
1035 EXPORT_SYMBOL(consume_skb);
1039 * __consume_stateless_skb - free an skbuff, assuming it is stateless
1040 * @skb: buffer to free
1042 * Alike consume_skb(), but this variant assumes that this is the last
1043 * skb reference and all the head states have been already dropped
1045 void __consume_stateless_skb(struct sk_buff *skb)
1047 trace_consume_skb(skb);
1048 skb_release_data(skb);
1052 static void napi_skb_cache_put(struct sk_buff *skb)
1054 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
1057 kasan_poison_object_data(skbuff_head_cache, skb);
1058 nc->skb_cache[nc->skb_count++] = skb;
1060 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
1061 for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++)
1062 kasan_unpoison_object_data(skbuff_head_cache,
1065 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_HALF,
1066 nc->skb_cache + NAPI_SKB_CACHE_HALF);
1067 nc->skb_count = NAPI_SKB_CACHE_HALF;
1071 void __kfree_skb_defer(struct sk_buff *skb)
1073 skb_release_all(skb);
1074 napi_skb_cache_put(skb);
1077 void napi_skb_free_stolen_head(struct sk_buff *skb)
1079 if (unlikely(skb->slow_gro)) {
1086 napi_skb_cache_put(skb);
1089 void napi_consume_skb(struct sk_buff *skb, int budget)
1091 /* Zero budget indicate non-NAPI context called us, like netpoll */
1092 if (unlikely(!budget)) {
1093 dev_consume_skb_any(skb);
1097 DEBUG_NET_WARN_ON_ONCE(!in_softirq());
1099 if (!skb_unref(skb))
1102 /* if reaching here SKB is ready to free */
1103 trace_consume_skb(skb);
1105 /* if SKB is a clone, don't handle this case */
1106 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
1111 skb_release_all(skb);
1112 napi_skb_cache_put(skb);
1114 EXPORT_SYMBOL(napi_consume_skb);
1116 /* Make sure a field is contained by headers group */
1117 #define CHECK_SKB_FIELD(field) \
1118 BUILD_BUG_ON(offsetof(struct sk_buff, field) != \
1119 offsetof(struct sk_buff, headers.field)); \
1121 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1123 new->tstamp = old->tstamp;
1124 /* We do not copy old->sk */
1125 new->dev = old->dev;
1126 memcpy(new->cb, old->cb, sizeof(old->cb));
1127 skb_dst_copy(new, old);
1128 __skb_ext_copy(new, old);
1129 __nf_copy(new, old, false);
1131 /* Note : this field could be in the headers group.
1132 * It is not yet because we do not want to have a 16 bit hole
1134 new->queue_mapping = old->queue_mapping;
1136 memcpy(&new->headers, &old->headers, sizeof(new->headers));
1137 CHECK_SKB_FIELD(protocol);
1138 CHECK_SKB_FIELD(csum);
1139 CHECK_SKB_FIELD(hash);
1140 CHECK_SKB_FIELD(priority);
1141 CHECK_SKB_FIELD(skb_iif);
1142 CHECK_SKB_FIELD(vlan_proto);
1143 CHECK_SKB_FIELD(vlan_tci);
1144 CHECK_SKB_FIELD(transport_header);
1145 CHECK_SKB_FIELD(network_header);
1146 CHECK_SKB_FIELD(mac_header);
1147 CHECK_SKB_FIELD(inner_protocol);
1148 CHECK_SKB_FIELD(inner_transport_header);
1149 CHECK_SKB_FIELD(inner_network_header);
1150 CHECK_SKB_FIELD(inner_mac_header);
1151 CHECK_SKB_FIELD(mark);
1152 #ifdef CONFIG_NETWORK_SECMARK
1153 CHECK_SKB_FIELD(secmark);
1155 #ifdef CONFIG_NET_RX_BUSY_POLL
1156 CHECK_SKB_FIELD(napi_id);
1158 CHECK_SKB_FIELD(alloc_cpu);
1160 CHECK_SKB_FIELD(sender_cpu);
1162 #ifdef CONFIG_NET_SCHED
1163 CHECK_SKB_FIELD(tc_index);
1169 * You should not add any new code to this function. Add it to
1170 * __copy_skb_header above instead.
1172 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
1174 #define C(x) n->x = skb->x
1176 n->next = n->prev = NULL;
1178 __copy_skb_header(n, skb);
1183 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
1189 n->destructor = NULL;
1196 refcount_set(&n->users, 1);
1198 atomic_inc(&(skb_shinfo(skb)->dataref));
1206 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1207 * @first: first sk_buff of the msg
1209 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1213 n = alloc_skb(0, GFP_ATOMIC);
1217 n->len = first->len;
1218 n->data_len = first->len;
1219 n->truesize = first->truesize;
1221 skb_shinfo(n)->frag_list = first;
1223 __copy_skb_header(n, first);
1224 n->destructor = NULL;
1228 EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1231 * skb_morph - morph one skb into another
1232 * @dst: the skb to receive the contents
1233 * @src: the skb to supply the contents
1235 * This is identical to skb_clone except that the target skb is
1236 * supplied by the user.
1238 * The target skb is returned upon exit.
1240 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1242 skb_release_all(dst);
1243 return __skb_clone(dst, src);
1245 EXPORT_SYMBOL_GPL(skb_morph);
1247 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1249 unsigned long max_pg, num_pg, new_pg, old_pg;
1250 struct user_struct *user;
1252 if (capable(CAP_IPC_LOCK) || !size)
1255 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
1256 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1257 user = mmp->user ? : current_user();
1260 old_pg = atomic_long_read(&user->locked_vm);
1261 new_pg = old_pg + num_pg;
1262 if (new_pg > max_pg)
1264 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
1268 mmp->user = get_uid(user);
1269 mmp->num_pg = num_pg;
1271 mmp->num_pg += num_pg;
1276 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1278 void mm_unaccount_pinned_pages(struct mmpin *mmp)
1281 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1282 free_uid(mmp->user);
1285 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1287 static struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size)
1289 struct ubuf_info_msgzc *uarg;
1290 struct sk_buff *skb;
1292 WARN_ON_ONCE(!in_task());
1294 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1298 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1299 uarg = (void *)skb->cb;
1300 uarg->mmp.user = NULL;
1302 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1307 uarg->ubuf.callback = msg_zerocopy_callback;
1308 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1310 uarg->bytelen = size;
1312 uarg->ubuf.flags = SKBFL_ZEROCOPY_FRAG | SKBFL_DONT_ORPHAN;
1313 refcount_set(&uarg->ubuf.refcnt, 1);
1319 static inline struct sk_buff *skb_from_uarg(struct ubuf_info_msgzc *uarg)
1321 return container_of((void *)uarg, struct sk_buff, cb);
1324 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1325 struct ubuf_info *uarg)
1328 struct ubuf_info_msgzc *uarg_zc;
1329 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
1332 /* there might be non MSG_ZEROCOPY users */
1333 if (uarg->callback != msg_zerocopy_callback)
1336 /* realloc only when socket is locked (TCP, UDP cork),
1337 * so uarg->len and sk_zckey access is serialized
1339 if (!sock_owned_by_user(sk)) {
1344 uarg_zc = uarg_to_msgzc(uarg);
1345 bytelen = uarg_zc->bytelen + size;
1346 if (uarg_zc->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1347 /* TCP can create new skb to attach new uarg */
1348 if (sk->sk_type == SOCK_STREAM)
1353 next = (u32)atomic_read(&sk->sk_zckey);
1354 if ((u32)(uarg_zc->id + uarg_zc->len) == next) {
1355 if (mm_account_pinned_pages(&uarg_zc->mmp, size))
1358 uarg_zc->bytelen = bytelen;
1359 atomic_set(&sk->sk_zckey, ++next);
1361 /* no extra ref when appending to datagram (MSG_MORE) */
1362 if (sk->sk_type == SOCK_STREAM)
1363 net_zcopy_get(uarg);
1370 return msg_zerocopy_alloc(sk, size);
1372 EXPORT_SYMBOL_GPL(msg_zerocopy_realloc);
1374 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1376 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1380 old_lo = serr->ee.ee_info;
1381 old_hi = serr->ee.ee_data;
1382 sum_len = old_hi - old_lo + 1ULL + len;
1384 if (sum_len >= (1ULL << 32))
1387 if (lo != old_hi + 1)
1390 serr->ee.ee_data += len;
1394 static void __msg_zerocopy_callback(struct ubuf_info_msgzc *uarg)
1396 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1397 struct sock_exterr_skb *serr;
1398 struct sock *sk = skb->sk;
1399 struct sk_buff_head *q;
1400 unsigned long flags;
1405 mm_unaccount_pinned_pages(&uarg->mmp);
1407 /* if !len, there was only 1 call, and it was aborted
1408 * so do not queue a completion notification
1410 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1415 hi = uarg->id + len - 1;
1416 is_zerocopy = uarg->zerocopy;
1418 serr = SKB_EXT_ERR(skb);
1419 memset(serr, 0, sizeof(*serr));
1420 serr->ee.ee_errno = 0;
1421 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1422 serr->ee.ee_data = hi;
1423 serr->ee.ee_info = lo;
1425 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1427 q = &sk->sk_error_queue;
1428 spin_lock_irqsave(&q->lock, flags);
1429 tail = skb_peek_tail(q);
1430 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1431 !skb_zerocopy_notify_extend(tail, lo, len)) {
1432 __skb_queue_tail(q, skb);
1435 spin_unlock_irqrestore(&q->lock, flags);
1437 sk_error_report(sk);
1444 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1447 struct ubuf_info_msgzc *uarg_zc = uarg_to_msgzc(uarg);
1449 uarg_zc->zerocopy = uarg_zc->zerocopy & success;
1451 if (refcount_dec_and_test(&uarg->refcnt))
1452 __msg_zerocopy_callback(uarg_zc);
1454 EXPORT_SYMBOL_GPL(msg_zerocopy_callback);
1456 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1458 struct sock *sk = skb_from_uarg(uarg_to_msgzc(uarg))->sk;
1460 atomic_dec(&sk->sk_zckey);
1461 uarg_to_msgzc(uarg)->len--;
1464 msg_zerocopy_callback(NULL, uarg, true);
1466 EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort);
1468 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1469 struct msghdr *msg, int len,
1470 struct ubuf_info *uarg)
1472 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1473 int err, orig_len = skb->len;
1475 /* An skb can only point to one uarg. This edge case happens when
1476 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1478 if (orig_uarg && uarg != orig_uarg)
1481 err = __zerocopy_sg_from_iter(msg, sk, skb, &msg->msg_iter, len);
1482 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1483 struct sock *save_sk = skb->sk;
1485 /* Streams do not free skb on error. Reset to prev state. */
1486 iov_iter_revert(&msg->msg_iter, skb->len - orig_len);
1488 ___pskb_trim(skb, orig_len);
1493 skb_zcopy_set(skb, uarg, NULL);
1494 return skb->len - orig_len;
1496 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1498 void __skb_zcopy_downgrade_managed(struct sk_buff *skb)
1502 skb_shinfo(skb)->flags &= ~SKBFL_MANAGED_FRAG_REFS;
1503 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1504 skb_frag_ref(skb, i);
1506 EXPORT_SYMBOL_GPL(__skb_zcopy_downgrade_managed);
1508 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1511 if (skb_zcopy(orig)) {
1512 if (skb_zcopy(nskb)) {
1513 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1518 if (skb_uarg(nskb) == skb_uarg(orig))
1520 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1523 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1529 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1530 * @skb: the skb to modify
1531 * @gfp_mask: allocation priority
1533 * This must be called on skb with SKBFL_ZEROCOPY_ENABLE.
1534 * It will copy all frags into kernel and drop the reference
1535 * to userspace pages.
1537 * If this function is called from an interrupt gfp_mask() must be
1540 * Returns 0 on success or a negative error code on failure
1541 * to allocate kernel memory to copy to.
1543 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1545 int num_frags = skb_shinfo(skb)->nr_frags;
1546 struct page *page, *head = NULL;
1550 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1556 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1557 for (i = 0; i < new_frags; i++) {
1558 page = alloc_page(gfp_mask);
1561 struct page *next = (struct page *)page_private(head);
1567 set_page_private(page, (unsigned long)head);
1573 for (i = 0; i < num_frags; i++) {
1574 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1575 u32 p_off, p_len, copied;
1579 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1580 p, p_off, p_len, copied) {
1582 vaddr = kmap_atomic(p);
1584 while (done < p_len) {
1585 if (d_off == PAGE_SIZE) {
1587 page = (struct page *)page_private(page);
1589 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1590 memcpy(page_address(page) + d_off,
1591 vaddr + p_off + done, copy);
1595 kunmap_atomic(vaddr);
1599 /* skb frags release userspace buffers */
1600 for (i = 0; i < num_frags; i++)
1601 skb_frag_unref(skb, i);
1603 /* skb frags point to kernel buffers */
1604 for (i = 0; i < new_frags - 1; i++) {
1605 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1606 head = (struct page *)page_private(head);
1608 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1609 skb_shinfo(skb)->nr_frags = new_frags;
1612 skb_zcopy_clear(skb, false);
1615 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1618 * skb_clone - duplicate an sk_buff
1619 * @skb: buffer to clone
1620 * @gfp_mask: allocation priority
1622 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1623 * copies share the same packet data but not structure. The new
1624 * buffer has a reference count of 1. If the allocation fails the
1625 * function returns %NULL otherwise the new buffer is returned.
1627 * If this function is called from an interrupt gfp_mask() must be
1631 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1633 struct sk_buff_fclones *fclones = container_of(skb,
1634 struct sk_buff_fclones,
1638 if (skb_orphan_frags(skb, gfp_mask))
1641 if (skb->fclone == SKB_FCLONE_ORIG &&
1642 refcount_read(&fclones->fclone_ref) == 1) {
1644 refcount_set(&fclones->fclone_ref, 2);
1645 n->fclone = SKB_FCLONE_CLONE;
1647 if (skb_pfmemalloc(skb))
1648 gfp_mask |= __GFP_MEMALLOC;
1650 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1654 n->fclone = SKB_FCLONE_UNAVAILABLE;
1657 return __skb_clone(n, skb);
1659 EXPORT_SYMBOL(skb_clone);
1661 void skb_headers_offset_update(struct sk_buff *skb, int off)
1663 /* Only adjust this if it actually is csum_start rather than csum */
1664 if (skb->ip_summed == CHECKSUM_PARTIAL)
1665 skb->csum_start += off;
1666 /* {transport,network,mac}_header and tail are relative to skb->head */
1667 skb->transport_header += off;
1668 skb->network_header += off;
1669 if (skb_mac_header_was_set(skb))
1670 skb->mac_header += off;
1671 skb->inner_transport_header += off;
1672 skb->inner_network_header += off;
1673 skb->inner_mac_header += off;
1675 EXPORT_SYMBOL(skb_headers_offset_update);
1677 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1679 __copy_skb_header(new, old);
1681 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1682 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1683 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1685 EXPORT_SYMBOL(skb_copy_header);
1687 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1689 if (skb_pfmemalloc(skb))
1690 return SKB_ALLOC_RX;
1695 * skb_copy - create private copy of an sk_buff
1696 * @skb: buffer to copy
1697 * @gfp_mask: allocation priority
1699 * Make a copy of both an &sk_buff and its data. This is used when the
1700 * caller wishes to modify the data and needs a private copy of the
1701 * data to alter. Returns %NULL on failure or the pointer to the buffer
1702 * on success. The returned buffer has a reference count of 1.
1704 * As by-product this function converts non-linear &sk_buff to linear
1705 * one, so that &sk_buff becomes completely private and caller is allowed
1706 * to modify all the data of returned buffer. This means that this
1707 * function is not recommended for use in circumstances when only
1708 * header is going to be modified. Use pskb_copy() instead.
1711 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1713 int headerlen = skb_headroom(skb);
1714 unsigned int size = skb_end_offset(skb) + skb->data_len;
1715 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1716 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1721 /* Set the data pointer */
1722 skb_reserve(n, headerlen);
1723 /* Set the tail pointer and length */
1724 skb_put(n, skb->len);
1726 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1728 skb_copy_header(n, skb);
1731 EXPORT_SYMBOL(skb_copy);
1734 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1735 * @skb: buffer to copy
1736 * @headroom: headroom of new skb
1737 * @gfp_mask: allocation priority
1738 * @fclone: if true allocate the copy of the skb from the fclone
1739 * cache instead of the head cache; it is recommended to set this
1740 * to true for the cases where the copy will likely be cloned
1742 * Make a copy of both an &sk_buff and part of its data, located
1743 * in header. Fragmented data remain shared. This is used when
1744 * the caller wishes to modify only header of &sk_buff and needs
1745 * private copy of the header to alter. Returns %NULL on failure
1746 * or the pointer to the buffer on success.
1747 * The returned buffer has a reference count of 1.
1750 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1751 gfp_t gfp_mask, bool fclone)
1753 unsigned int size = skb_headlen(skb) + headroom;
1754 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1755 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1760 /* Set the data pointer */
1761 skb_reserve(n, headroom);
1762 /* Set the tail pointer and length */
1763 skb_put(n, skb_headlen(skb));
1764 /* Copy the bytes */
1765 skb_copy_from_linear_data(skb, n->data, n->len);
1767 n->truesize += skb->data_len;
1768 n->data_len = skb->data_len;
1771 if (skb_shinfo(skb)->nr_frags) {
1774 if (skb_orphan_frags(skb, gfp_mask) ||
1775 skb_zerocopy_clone(n, skb, gfp_mask)) {
1780 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1781 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1782 skb_frag_ref(skb, i);
1784 skb_shinfo(n)->nr_frags = i;
1787 if (skb_has_frag_list(skb)) {
1788 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1789 skb_clone_fraglist(n);
1792 skb_copy_header(n, skb);
1796 EXPORT_SYMBOL(__pskb_copy_fclone);
1799 * pskb_expand_head - reallocate header of &sk_buff
1800 * @skb: buffer to reallocate
1801 * @nhead: room to add at head
1802 * @ntail: room to add at tail
1803 * @gfp_mask: allocation priority
1805 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1806 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1807 * reference count of 1. Returns zero in the case of success or error,
1808 * if expansion failed. In the last case, &sk_buff is not changed.
1810 * All the pointers pointing into skb header may change and must be
1811 * reloaded after call to this function.
1814 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1817 int i, osize = skb_end_offset(skb);
1818 int size = osize + nhead + ntail;
1824 BUG_ON(skb_shared(skb));
1826 skb_zcopy_downgrade_managed(skb);
1828 size = SKB_DATA_ALIGN(size);
1830 if (skb_pfmemalloc(skb))
1831 gfp_mask |= __GFP_MEMALLOC;
1832 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1833 gfp_mask, NUMA_NO_NODE, NULL);
1836 size = SKB_WITH_OVERHEAD(ksize(data));
1838 /* Copy only real data... and, alas, header. This should be
1839 * optimized for the cases when header is void.
1841 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1843 memcpy((struct skb_shared_info *)(data + size),
1845 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1848 * if shinfo is shared we must drop the old head gracefully, but if it
1849 * is not we can just drop the old head and let the existing refcount
1850 * be since all we did is relocate the values
1852 if (skb_cloned(skb)) {
1853 if (skb_orphan_frags(skb, gfp_mask))
1856 refcount_inc(&skb_uarg(skb)->refcnt);
1857 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1858 skb_frag_ref(skb, i);
1860 if (skb_has_frag_list(skb))
1861 skb_clone_fraglist(skb);
1863 skb_release_data(skb);
1867 off = (data + nhead) - skb->head;
1873 skb_set_end_offset(skb, size);
1874 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1878 skb_headers_offset_update(skb, nhead);
1882 atomic_set(&skb_shinfo(skb)->dataref, 1);
1884 skb_metadata_clear(skb);
1886 /* It is not generally safe to change skb->truesize.
1887 * For the moment, we really care of rx path, or
1888 * when skb is orphaned (not attached to a socket).
1890 if (!skb->sk || skb->destructor == sock_edemux)
1891 skb->truesize += size - osize;
1900 EXPORT_SYMBOL(pskb_expand_head);
1902 /* Make private copy of skb with writable head and some headroom */
1904 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1906 struct sk_buff *skb2;
1907 int delta = headroom - skb_headroom(skb);
1910 skb2 = pskb_copy(skb, GFP_ATOMIC);
1912 skb2 = skb_clone(skb, GFP_ATOMIC);
1913 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1921 EXPORT_SYMBOL(skb_realloc_headroom);
1923 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
1925 unsigned int saved_end_offset, saved_truesize;
1926 struct skb_shared_info *shinfo;
1929 saved_end_offset = skb_end_offset(skb);
1930 saved_truesize = skb->truesize;
1932 res = pskb_expand_head(skb, 0, 0, pri);
1936 skb->truesize = saved_truesize;
1938 if (likely(skb_end_offset(skb) == saved_end_offset))
1941 shinfo = skb_shinfo(skb);
1943 /* We are about to change back skb->end,
1944 * we need to move skb_shinfo() to its new location.
1946 memmove(skb->head + saved_end_offset,
1948 offsetof(struct skb_shared_info, frags[shinfo->nr_frags]));
1950 skb_set_end_offset(skb, saved_end_offset);
1956 * skb_expand_head - reallocate header of &sk_buff
1957 * @skb: buffer to reallocate
1958 * @headroom: needed headroom
1960 * Unlike skb_realloc_headroom, this one does not allocate a new skb
1961 * if possible; copies skb->sk to new skb as needed
1962 * and frees original skb in case of failures.
1964 * It expect increased headroom and generates warning otherwise.
1967 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom)
1969 int delta = headroom - skb_headroom(skb);
1970 int osize = skb_end_offset(skb);
1971 struct sock *sk = skb->sk;
1973 if (WARN_ONCE(delta <= 0,
1974 "%s is expecting an increase in the headroom", __func__))
1977 delta = SKB_DATA_ALIGN(delta);
1978 /* pskb_expand_head() might crash, if skb is shared. */
1979 if (skb_shared(skb) || !is_skb_wmem(skb)) {
1980 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1982 if (unlikely(!nskb))
1986 skb_set_owner_w(nskb, sk);
1990 if (pskb_expand_head(skb, delta, 0, GFP_ATOMIC))
1993 if (sk && is_skb_wmem(skb)) {
1994 delta = skb_end_offset(skb) - osize;
1995 refcount_add(delta, &sk->sk_wmem_alloc);
1996 skb->truesize += delta;
2004 EXPORT_SYMBOL(skb_expand_head);
2007 * skb_copy_expand - copy and expand sk_buff
2008 * @skb: buffer to copy
2009 * @newheadroom: new free bytes at head
2010 * @newtailroom: new free bytes at tail
2011 * @gfp_mask: allocation priority
2013 * Make a copy of both an &sk_buff and its data and while doing so
2014 * allocate additional space.
2016 * This is used when the caller wishes to modify the data and needs a
2017 * private copy of the data to alter as well as more space for new fields.
2018 * Returns %NULL on failure or the pointer to the buffer
2019 * on success. The returned buffer has a reference count of 1.
2021 * You must pass %GFP_ATOMIC as the allocation priority if this function
2022 * is called from an interrupt.
2024 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
2025 int newheadroom, int newtailroom,
2029 * Allocate the copy buffer
2031 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
2032 gfp_mask, skb_alloc_rx_flag(skb),
2034 int oldheadroom = skb_headroom(skb);
2035 int head_copy_len, head_copy_off;
2040 skb_reserve(n, newheadroom);
2042 /* Set the tail pointer and length */
2043 skb_put(n, skb->len);
2045 head_copy_len = oldheadroom;
2047 if (newheadroom <= head_copy_len)
2048 head_copy_len = newheadroom;
2050 head_copy_off = newheadroom - head_copy_len;
2052 /* Copy the linear header and data. */
2053 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
2054 skb->len + head_copy_len));
2056 skb_copy_header(n, skb);
2058 skb_headers_offset_update(n, newheadroom - oldheadroom);
2062 EXPORT_SYMBOL(skb_copy_expand);
2065 * __skb_pad - zero pad the tail of an skb
2066 * @skb: buffer to pad
2067 * @pad: space to pad
2068 * @free_on_error: free buffer on error
2070 * Ensure that a buffer is followed by a padding area that is zero
2071 * filled. Used by network drivers which may DMA or transfer data
2072 * beyond the buffer end onto the wire.
2074 * May return error in out of memory cases. The skb is freed on error
2075 * if @free_on_error is true.
2078 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
2083 /* If the skbuff is non linear tailroom is always zero.. */
2084 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
2085 memset(skb->data+skb->len, 0, pad);
2089 ntail = skb->data_len + pad - (skb->end - skb->tail);
2090 if (likely(skb_cloned(skb) || ntail > 0)) {
2091 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
2096 /* FIXME: The use of this function with non-linear skb's really needs
2099 err = skb_linearize(skb);
2103 memset(skb->data + skb->len, 0, pad);
2111 EXPORT_SYMBOL(__skb_pad);
2114 * pskb_put - add data to the tail of a potentially fragmented buffer
2115 * @skb: start of the buffer to use
2116 * @tail: tail fragment of the buffer to use
2117 * @len: amount of data to add
2119 * This function extends the used data area of the potentially
2120 * fragmented buffer. @tail must be the last fragment of @skb -- or
2121 * @skb itself. If this would exceed the total buffer size the kernel
2122 * will panic. A pointer to the first byte of the extra data is
2126 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
2129 skb->data_len += len;
2132 return skb_put(tail, len);
2134 EXPORT_SYMBOL_GPL(pskb_put);
2137 * skb_put - add data to a buffer
2138 * @skb: buffer to use
2139 * @len: amount of data to add
2141 * This function extends the used data area of the buffer. If this would
2142 * exceed the total buffer size the kernel will panic. A pointer to the
2143 * first byte of the extra data is returned.
2145 void *skb_put(struct sk_buff *skb, unsigned int len)
2147 void *tmp = skb_tail_pointer(skb);
2148 SKB_LINEAR_ASSERT(skb);
2151 if (unlikely(skb->tail > skb->end))
2152 skb_over_panic(skb, len, __builtin_return_address(0));
2155 EXPORT_SYMBOL(skb_put);
2158 * skb_push - add data to the start of a buffer
2159 * @skb: buffer to use
2160 * @len: amount of data to add
2162 * This function extends the used data area of the buffer at the buffer
2163 * start. If this would exceed the total buffer headroom the kernel will
2164 * panic. A pointer to the first byte of the extra data is returned.
2166 void *skb_push(struct sk_buff *skb, unsigned int len)
2170 if (unlikely(skb->data < skb->head))
2171 skb_under_panic(skb, len, __builtin_return_address(0));
2174 EXPORT_SYMBOL(skb_push);
2177 * skb_pull - remove data from the start of a buffer
2178 * @skb: buffer to use
2179 * @len: amount of data to remove
2181 * This function removes data from the start of a buffer, returning
2182 * the memory to the headroom. A pointer to the next data in the buffer
2183 * is returned. Once the data has been pulled future pushes will overwrite
2186 void *skb_pull(struct sk_buff *skb, unsigned int len)
2188 return skb_pull_inline(skb, len);
2190 EXPORT_SYMBOL(skb_pull);
2193 * skb_pull_data - remove data from the start of a buffer returning its
2194 * original position.
2195 * @skb: buffer to use
2196 * @len: amount of data to remove
2198 * This function removes data from the start of a buffer, returning
2199 * the memory to the headroom. A pointer to the original data in the buffer
2200 * is returned after checking if there is enough data to pull. Once the
2201 * data has been pulled future pushes will overwrite the old data.
2203 void *skb_pull_data(struct sk_buff *skb, size_t len)
2205 void *data = skb->data;
2214 EXPORT_SYMBOL(skb_pull_data);
2217 * skb_trim - remove end from a buffer
2218 * @skb: buffer to alter
2221 * Cut the length of a buffer down by removing data from the tail. If
2222 * the buffer is already under the length specified it is not modified.
2223 * The skb must be linear.
2225 void skb_trim(struct sk_buff *skb, unsigned int len)
2228 __skb_trim(skb, len);
2230 EXPORT_SYMBOL(skb_trim);
2232 /* Trims skb to length len. It can change skb pointers.
2235 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
2237 struct sk_buff **fragp;
2238 struct sk_buff *frag;
2239 int offset = skb_headlen(skb);
2240 int nfrags = skb_shinfo(skb)->nr_frags;
2244 if (skb_cloned(skb) &&
2245 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
2252 for (; i < nfrags; i++) {
2253 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2260 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
2263 skb_shinfo(skb)->nr_frags = i;
2265 for (; i < nfrags; i++)
2266 skb_frag_unref(skb, i);
2268 if (skb_has_frag_list(skb))
2269 skb_drop_fraglist(skb);
2273 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
2274 fragp = &frag->next) {
2275 int end = offset + frag->len;
2277 if (skb_shared(frag)) {
2278 struct sk_buff *nfrag;
2280 nfrag = skb_clone(frag, GFP_ATOMIC);
2281 if (unlikely(!nfrag))
2284 nfrag->next = frag->next;
2296 unlikely((err = pskb_trim(frag, len - offset))))
2300 skb_drop_list(&frag->next);
2305 if (len > skb_headlen(skb)) {
2306 skb->data_len -= skb->len - len;
2311 skb_set_tail_pointer(skb, len);
2314 if (!skb->sk || skb->destructor == sock_edemux)
2318 EXPORT_SYMBOL(___pskb_trim);
2320 /* Note : use pskb_trim_rcsum() instead of calling this directly
2322 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2324 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2325 int delta = skb->len - len;
2327 skb->csum = csum_block_sub(skb->csum,
2328 skb_checksum(skb, len, delta, 0),
2330 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
2331 int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len;
2332 int offset = skb_checksum_start_offset(skb) + skb->csum_offset;
2334 if (offset + sizeof(__sum16) > hdlen)
2337 return __pskb_trim(skb, len);
2339 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2342 * __pskb_pull_tail - advance tail of skb header
2343 * @skb: buffer to reallocate
2344 * @delta: number of bytes to advance tail
2346 * The function makes a sense only on a fragmented &sk_buff,
2347 * it expands header moving its tail forward and copying necessary
2348 * data from fragmented part.
2350 * &sk_buff MUST have reference count of 1.
2352 * Returns %NULL (and &sk_buff does not change) if pull failed
2353 * or value of new tail of skb in the case of success.
2355 * All the pointers pointing into skb header may change and must be
2356 * reloaded after call to this function.
2359 /* Moves tail of skb head forward, copying data from fragmented part,
2360 * when it is necessary.
2361 * 1. It may fail due to malloc failure.
2362 * 2. It may change skb pointers.
2364 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2366 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2368 /* If skb has not enough free space at tail, get new one
2369 * plus 128 bytes for future expansions. If we have enough
2370 * room at tail, reallocate without expansion only if skb is cloned.
2372 int i, k, eat = (skb->tail + delta) - skb->end;
2374 if (eat > 0 || skb_cloned(skb)) {
2375 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2380 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2381 skb_tail_pointer(skb), delta));
2383 /* Optimization: no fragments, no reasons to preestimate
2384 * size of pulled pages. Superb.
2386 if (!skb_has_frag_list(skb))
2389 /* Estimate size of pulled pages. */
2391 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2392 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2399 /* If we need update frag list, we are in troubles.
2400 * Certainly, it is possible to add an offset to skb data,
2401 * but taking into account that pulling is expected to
2402 * be very rare operation, it is worth to fight against
2403 * further bloating skb head and crucify ourselves here instead.
2404 * Pure masohism, indeed. 8)8)
2407 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2408 struct sk_buff *clone = NULL;
2409 struct sk_buff *insp = NULL;
2412 if (list->len <= eat) {
2413 /* Eaten as whole. */
2418 /* Eaten partially. */
2419 if (skb_is_gso(skb) && !list->head_frag &&
2421 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2423 if (skb_shared(list)) {
2424 /* Sucks! We need to fork list. :-( */
2425 clone = skb_clone(list, GFP_ATOMIC);
2431 /* This may be pulled without
2435 if (!pskb_pull(list, eat)) {
2443 /* Free pulled out fragments. */
2444 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2445 skb_shinfo(skb)->frag_list = list->next;
2448 /* And insert new clone at head. */
2451 skb_shinfo(skb)->frag_list = clone;
2454 /* Success! Now we may commit changes to skb data. */
2459 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2460 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2463 skb_frag_unref(skb, i);
2466 skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2468 *frag = skb_shinfo(skb)->frags[i];
2470 skb_frag_off_add(frag, eat);
2471 skb_frag_size_sub(frag, eat);
2479 skb_shinfo(skb)->nr_frags = k;
2483 skb->data_len -= delta;
2486 skb_zcopy_clear(skb, false);
2488 return skb_tail_pointer(skb);
2490 EXPORT_SYMBOL(__pskb_pull_tail);
2493 * skb_copy_bits - copy bits from skb to kernel buffer
2495 * @offset: offset in source
2496 * @to: destination buffer
2497 * @len: number of bytes to copy
2499 * Copy the specified number of bytes from the source skb to the
2500 * destination buffer.
2503 * If its prototype is ever changed,
2504 * check arch/{*}/net/{*}.S files,
2505 * since it is called from BPF assembly code.
2507 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2509 int start = skb_headlen(skb);
2510 struct sk_buff *frag_iter;
2513 if (offset > (int)skb->len - len)
2517 if ((copy = start - offset) > 0) {
2520 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2521 if ((len -= copy) == 0)
2527 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2529 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2531 WARN_ON(start > offset + len);
2533 end = start + skb_frag_size(f);
2534 if ((copy = end - offset) > 0) {
2535 u32 p_off, p_len, copied;
2542 skb_frag_foreach_page(f,
2543 skb_frag_off(f) + offset - start,
2544 copy, p, p_off, p_len, copied) {
2545 vaddr = kmap_atomic(p);
2546 memcpy(to + copied, vaddr + p_off, p_len);
2547 kunmap_atomic(vaddr);
2550 if ((len -= copy) == 0)
2558 skb_walk_frags(skb, frag_iter) {
2561 WARN_ON(start > offset + len);
2563 end = start + frag_iter->len;
2564 if ((copy = end - offset) > 0) {
2567 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2569 if ((len -= copy) == 0)
2583 EXPORT_SYMBOL(skb_copy_bits);
2586 * Callback from splice_to_pipe(), if we need to release some pages
2587 * at the end of the spd in case we error'ed out in filling the pipe.
2589 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2591 put_page(spd->pages[i]);
2594 static struct page *linear_to_page(struct page *page, unsigned int *len,
2595 unsigned int *offset,
2598 struct page_frag *pfrag = sk_page_frag(sk);
2600 if (!sk_page_frag_refill(sk, pfrag))
2603 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2605 memcpy(page_address(pfrag->page) + pfrag->offset,
2606 page_address(page) + *offset, *len);
2607 *offset = pfrag->offset;
2608 pfrag->offset += *len;
2613 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2615 unsigned int offset)
2617 return spd->nr_pages &&
2618 spd->pages[spd->nr_pages - 1] == page &&
2619 (spd->partial[spd->nr_pages - 1].offset +
2620 spd->partial[spd->nr_pages - 1].len == offset);
2624 * Fill page/offset/length into spd, if it can hold more pages.
2626 static bool spd_fill_page(struct splice_pipe_desc *spd,
2627 struct pipe_inode_info *pipe, struct page *page,
2628 unsigned int *len, unsigned int offset,
2632 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2636 page = linear_to_page(page, len, &offset, sk);
2640 if (spd_can_coalesce(spd, page, offset)) {
2641 spd->partial[spd->nr_pages - 1].len += *len;
2645 spd->pages[spd->nr_pages] = page;
2646 spd->partial[spd->nr_pages].len = *len;
2647 spd->partial[spd->nr_pages].offset = offset;
2653 static bool __splice_segment(struct page *page, unsigned int poff,
2654 unsigned int plen, unsigned int *off,
2656 struct splice_pipe_desc *spd, bool linear,
2658 struct pipe_inode_info *pipe)
2663 /* skip this segment if already processed */
2669 /* ignore any bits we already processed */
2675 unsigned int flen = min(*len, plen);
2677 if (spd_fill_page(spd, pipe, page, &flen, poff,
2683 } while (*len && plen);
2689 * Map linear and fragment data from the skb to spd. It reports true if the
2690 * pipe is full or if we already spliced the requested length.
2692 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2693 unsigned int *offset, unsigned int *len,
2694 struct splice_pipe_desc *spd, struct sock *sk)
2697 struct sk_buff *iter;
2699 /* map the linear part :
2700 * If skb->head_frag is set, this 'linear' part is backed by a
2701 * fragment, and if the head is not shared with any clones then
2702 * we can avoid a copy since we own the head portion of this page.
2704 if (__splice_segment(virt_to_page(skb->data),
2705 (unsigned long) skb->data & (PAGE_SIZE - 1),
2708 skb_head_is_locked(skb),
2713 * then map the fragments
2715 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2716 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2718 if (__splice_segment(skb_frag_page(f),
2719 skb_frag_off(f), skb_frag_size(f),
2720 offset, len, spd, false, sk, pipe))
2724 skb_walk_frags(skb, iter) {
2725 if (*offset >= iter->len) {
2726 *offset -= iter->len;
2729 /* __skb_splice_bits() only fails if the output has no room
2730 * left, so no point in going over the frag_list for the error
2733 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2741 * Map data from the skb to a pipe. Should handle both the linear part,
2742 * the fragments, and the frag list.
2744 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2745 struct pipe_inode_info *pipe, unsigned int tlen,
2748 struct partial_page partial[MAX_SKB_FRAGS];
2749 struct page *pages[MAX_SKB_FRAGS];
2750 struct splice_pipe_desc spd = {
2753 .nr_pages_max = MAX_SKB_FRAGS,
2754 .ops = &nosteal_pipe_buf_ops,
2755 .spd_release = sock_spd_release,
2759 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2762 ret = splice_to_pipe(pipe, &spd);
2766 EXPORT_SYMBOL_GPL(skb_splice_bits);
2768 static int sendmsg_unlocked(struct sock *sk, struct msghdr *msg,
2769 struct kvec *vec, size_t num, size_t size)
2771 struct socket *sock = sk->sk_socket;
2775 return kernel_sendmsg(sock, msg, vec, num, size);
2778 static int sendpage_unlocked(struct sock *sk, struct page *page, int offset,
2779 size_t size, int flags)
2781 struct socket *sock = sk->sk_socket;
2785 return kernel_sendpage(sock, page, offset, size, flags);
2788 typedef int (*sendmsg_func)(struct sock *sk, struct msghdr *msg,
2789 struct kvec *vec, size_t num, size_t size);
2790 typedef int (*sendpage_func)(struct sock *sk, struct page *page, int offset,
2791 size_t size, int flags);
2792 static int __skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset,
2793 int len, sendmsg_func sendmsg, sendpage_func sendpage)
2795 unsigned int orig_len = len;
2796 struct sk_buff *head = skb;
2797 unsigned short fragidx;
2802 /* Deal with head data */
2803 while (offset < skb_headlen(skb) && len) {
2807 slen = min_t(int, len, skb_headlen(skb) - offset);
2808 kv.iov_base = skb->data + offset;
2810 memset(&msg, 0, sizeof(msg));
2811 msg.msg_flags = MSG_DONTWAIT;
2813 ret = INDIRECT_CALL_2(sendmsg, kernel_sendmsg_locked,
2814 sendmsg_unlocked, sk, &msg, &kv, 1, slen);
2822 /* All the data was skb head? */
2826 /* Make offset relative to start of frags */
2827 offset -= skb_headlen(skb);
2829 /* Find where we are in frag list */
2830 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2831 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2833 if (offset < skb_frag_size(frag))
2836 offset -= skb_frag_size(frag);
2839 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2840 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2842 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2845 ret = INDIRECT_CALL_2(sendpage, kernel_sendpage_locked,
2846 sendpage_unlocked, sk,
2847 skb_frag_page(frag),
2848 skb_frag_off(frag) + offset,
2849 slen, MSG_DONTWAIT);
2862 /* Process any frag lists */
2865 if (skb_has_frag_list(skb)) {
2866 skb = skb_shinfo(skb)->frag_list;
2869 } else if (skb->next) {
2876 return orig_len - len;
2879 return orig_len == len ? ret : orig_len - len;
2882 /* Send skb data on a socket. Socket must be locked. */
2883 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2886 return __skb_send_sock(sk, skb, offset, len, kernel_sendmsg_locked,
2887 kernel_sendpage_locked);
2889 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2891 /* Send skb data on a socket. Socket must be unlocked. */
2892 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len)
2894 return __skb_send_sock(sk, skb, offset, len, sendmsg_unlocked,
2899 * skb_store_bits - store bits from kernel buffer to skb
2900 * @skb: destination buffer
2901 * @offset: offset in destination
2902 * @from: source buffer
2903 * @len: number of bytes to copy
2905 * Copy the specified number of bytes from the source buffer to the
2906 * destination skb. This function handles all the messy bits of
2907 * traversing fragment lists and such.
2910 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2912 int start = skb_headlen(skb);
2913 struct sk_buff *frag_iter;
2916 if (offset > (int)skb->len - len)
2919 if ((copy = start - offset) > 0) {
2922 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2923 if ((len -= copy) == 0)
2929 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2930 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2933 WARN_ON(start > offset + len);
2935 end = start + skb_frag_size(frag);
2936 if ((copy = end - offset) > 0) {
2937 u32 p_off, p_len, copied;
2944 skb_frag_foreach_page(frag,
2945 skb_frag_off(frag) + offset - start,
2946 copy, p, p_off, p_len, copied) {
2947 vaddr = kmap_atomic(p);
2948 memcpy(vaddr + p_off, from + copied, p_len);
2949 kunmap_atomic(vaddr);
2952 if ((len -= copy) == 0)
2960 skb_walk_frags(skb, frag_iter) {
2963 WARN_ON(start > offset + len);
2965 end = start + frag_iter->len;
2966 if ((copy = end - offset) > 0) {
2969 if (skb_store_bits(frag_iter, offset - start,
2972 if ((len -= copy) == 0)
2985 EXPORT_SYMBOL(skb_store_bits);
2987 /* Checksum skb data. */
2988 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2989 __wsum csum, const struct skb_checksum_ops *ops)
2991 int start = skb_headlen(skb);
2992 int i, copy = start - offset;
2993 struct sk_buff *frag_iter;
2996 /* Checksum header. */
3000 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
3001 skb->data + offset, copy, csum);
3002 if ((len -= copy) == 0)
3008 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3010 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3012 WARN_ON(start > offset + len);
3014 end = start + skb_frag_size(frag);
3015 if ((copy = end - offset) > 0) {
3016 u32 p_off, p_len, copied;
3024 skb_frag_foreach_page(frag,
3025 skb_frag_off(frag) + offset - start,
3026 copy, p, p_off, p_len, copied) {
3027 vaddr = kmap_atomic(p);
3028 csum2 = INDIRECT_CALL_1(ops->update,
3030 vaddr + p_off, p_len, 0);
3031 kunmap_atomic(vaddr);
3032 csum = INDIRECT_CALL_1(ops->combine,
3033 csum_block_add_ext, csum,
3045 skb_walk_frags(skb, frag_iter) {
3048 WARN_ON(start > offset + len);
3050 end = start + frag_iter->len;
3051 if ((copy = end - offset) > 0) {
3055 csum2 = __skb_checksum(frag_iter, offset - start,
3057 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
3058 csum, csum2, pos, copy);
3059 if ((len -= copy) == 0)
3070 EXPORT_SYMBOL(__skb_checksum);
3072 __wsum skb_checksum(const struct sk_buff *skb, int offset,
3073 int len, __wsum csum)
3075 const struct skb_checksum_ops ops = {
3076 .update = csum_partial_ext,
3077 .combine = csum_block_add_ext,
3080 return __skb_checksum(skb, offset, len, csum, &ops);
3082 EXPORT_SYMBOL(skb_checksum);
3084 /* Both of above in one bottle. */
3086 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
3089 int start = skb_headlen(skb);
3090 int i, copy = start - offset;
3091 struct sk_buff *frag_iter;
3099 csum = csum_partial_copy_nocheck(skb->data + offset, to,
3101 if ((len -= copy) == 0)
3108 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3111 WARN_ON(start > offset + len);
3113 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3114 if ((copy = end - offset) > 0) {
3115 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3116 u32 p_off, p_len, copied;
3124 skb_frag_foreach_page(frag,
3125 skb_frag_off(frag) + offset - start,
3126 copy, p, p_off, p_len, copied) {
3127 vaddr = kmap_atomic(p);
3128 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
3131 kunmap_atomic(vaddr);
3132 csum = csum_block_add(csum, csum2, pos);
3144 skb_walk_frags(skb, frag_iter) {
3148 WARN_ON(start > offset + len);
3150 end = start + frag_iter->len;
3151 if ((copy = end - offset) > 0) {
3154 csum2 = skb_copy_and_csum_bits(frag_iter,
3157 csum = csum_block_add(csum, csum2, pos);
3158 if ((len -= copy) == 0)
3169 EXPORT_SYMBOL(skb_copy_and_csum_bits);
3171 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
3175 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
3176 /* See comments in __skb_checksum_complete(). */
3178 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
3179 !skb->csum_complete_sw)
3180 netdev_rx_csum_fault(skb->dev, skb);
3182 if (!skb_shared(skb))
3183 skb->csum_valid = !sum;
3186 EXPORT_SYMBOL(__skb_checksum_complete_head);
3188 /* This function assumes skb->csum already holds pseudo header's checksum,
3189 * which has been changed from the hardware checksum, for example, by
3190 * __skb_checksum_validate_complete(). And, the original skb->csum must
3191 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
3193 * It returns non-zero if the recomputed checksum is still invalid, otherwise
3194 * zero. The new checksum is stored back into skb->csum unless the skb is
3197 __sum16 __skb_checksum_complete(struct sk_buff *skb)
3202 csum = skb_checksum(skb, 0, skb->len, 0);
3204 sum = csum_fold(csum_add(skb->csum, csum));
3205 /* This check is inverted, because we already knew the hardware
3206 * checksum is invalid before calling this function. So, if the
3207 * re-computed checksum is valid instead, then we have a mismatch
3208 * between the original skb->csum and skb_checksum(). This means either
3209 * the original hardware checksum is incorrect or we screw up skb->csum
3210 * when moving skb->data around.
3213 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
3214 !skb->csum_complete_sw)
3215 netdev_rx_csum_fault(skb->dev, skb);
3218 if (!skb_shared(skb)) {
3219 /* Save full packet checksum */
3221 skb->ip_summed = CHECKSUM_COMPLETE;
3222 skb->csum_complete_sw = 1;
3223 skb->csum_valid = !sum;
3228 EXPORT_SYMBOL(__skb_checksum_complete);
3230 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
3232 net_warn_ratelimited(
3233 "%s: attempt to compute crc32c without libcrc32c.ko\n",
3238 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
3239 int offset, int len)
3241 net_warn_ratelimited(
3242 "%s: attempt to compute crc32c without libcrc32c.ko\n",
3247 static const struct skb_checksum_ops default_crc32c_ops = {
3248 .update = warn_crc32c_csum_update,
3249 .combine = warn_crc32c_csum_combine,
3252 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
3253 &default_crc32c_ops;
3254 EXPORT_SYMBOL(crc32c_csum_stub);
3257 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
3258 * @from: source buffer
3260 * Calculates the amount of linear headroom needed in the 'to' skb passed
3261 * into skb_zerocopy().
3264 skb_zerocopy_headlen(const struct sk_buff *from)
3266 unsigned int hlen = 0;
3268 if (!from->head_frag ||
3269 skb_headlen(from) < L1_CACHE_BYTES ||
3270 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) {
3271 hlen = skb_headlen(from);
3276 if (skb_has_frag_list(from))
3281 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
3284 * skb_zerocopy - Zero copy skb to skb
3285 * @to: destination buffer
3286 * @from: source buffer
3287 * @len: number of bytes to copy from source buffer
3288 * @hlen: size of linear headroom in destination buffer
3290 * Copies up to `len` bytes from `from` to `to` by creating references
3291 * to the frags in the source buffer.
3293 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
3294 * headroom in the `to` buffer.
3297 * 0: everything is OK
3298 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
3299 * -EFAULT: skb_copy_bits() found some problem with skb geometry
3302 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
3305 int plen = 0; /* length of skb->head fragment */
3308 unsigned int offset;
3310 BUG_ON(!from->head_frag && !hlen);
3312 /* dont bother with small payloads */
3313 if (len <= skb_tailroom(to))
3314 return skb_copy_bits(from, 0, skb_put(to, len), len);
3317 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
3322 plen = min_t(int, skb_headlen(from), len);
3324 page = virt_to_head_page(from->head);
3325 offset = from->data - (unsigned char *)page_address(page);
3326 __skb_fill_page_desc(to, 0, page, offset, plen);
3333 skb_len_add(to, len + plen);
3335 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
3339 skb_zerocopy_clone(to, from, GFP_ATOMIC);
3341 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
3346 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
3347 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
3349 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
3351 skb_frag_ref(to, j);
3354 skb_shinfo(to)->nr_frags = j;
3358 EXPORT_SYMBOL_GPL(skb_zerocopy);
3360 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3365 if (skb->ip_summed == CHECKSUM_PARTIAL)
3366 csstart = skb_checksum_start_offset(skb);
3368 csstart = skb_headlen(skb);
3370 BUG_ON(csstart > skb_headlen(skb));
3372 skb_copy_from_linear_data(skb, to, csstart);
3375 if (csstart != skb->len)
3376 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3377 skb->len - csstart);
3379 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3380 long csstuff = csstart + skb->csum_offset;
3382 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3385 EXPORT_SYMBOL(skb_copy_and_csum_dev);
3388 * skb_dequeue - remove from the head of the queue
3389 * @list: list to dequeue from
3391 * Remove the head of the list. The list lock is taken so the function
3392 * may be used safely with other locking list functions. The head item is
3393 * returned or %NULL if the list is empty.
3396 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3398 unsigned long flags;
3399 struct sk_buff *result;
3401 spin_lock_irqsave(&list->lock, flags);
3402 result = __skb_dequeue(list);
3403 spin_unlock_irqrestore(&list->lock, flags);
3406 EXPORT_SYMBOL(skb_dequeue);
3409 * skb_dequeue_tail - remove from the tail of the queue
3410 * @list: list to dequeue from
3412 * Remove the tail of the list. The list lock is taken so the function
3413 * may be used safely with other locking list functions. The tail item is
3414 * returned or %NULL if the list is empty.
3416 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3418 unsigned long flags;
3419 struct sk_buff *result;
3421 spin_lock_irqsave(&list->lock, flags);
3422 result = __skb_dequeue_tail(list);
3423 spin_unlock_irqrestore(&list->lock, flags);
3426 EXPORT_SYMBOL(skb_dequeue_tail);
3429 * skb_queue_purge - empty a list
3430 * @list: list to empty
3432 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3433 * the list and one reference dropped. This function takes the list
3434 * lock and is atomic with respect to other list locking functions.
3436 void skb_queue_purge(struct sk_buff_head *list)
3438 struct sk_buff *skb;
3439 while ((skb = skb_dequeue(list)) != NULL)
3442 EXPORT_SYMBOL(skb_queue_purge);
3445 * skb_rbtree_purge - empty a skb rbtree
3446 * @root: root of the rbtree to empty
3447 * Return value: the sum of truesizes of all purged skbs.
3449 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3450 * the list and one reference dropped. This function does not take
3451 * any lock. Synchronization should be handled by the caller (e.g., TCP
3452 * out-of-order queue is protected by the socket lock).
3454 unsigned int skb_rbtree_purge(struct rb_root *root)
3456 struct rb_node *p = rb_first(root);
3457 unsigned int sum = 0;
3460 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3463 rb_erase(&skb->rbnode, root);
3464 sum += skb->truesize;
3471 * skb_queue_head - queue a buffer at the list head
3472 * @list: list to use
3473 * @newsk: buffer to queue
3475 * Queue a buffer at the start of the list. This function takes the
3476 * list lock and can be used safely with other locking &sk_buff functions
3479 * A buffer cannot be placed on two lists at the same time.
3481 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3483 unsigned long flags;
3485 spin_lock_irqsave(&list->lock, flags);
3486 __skb_queue_head(list, newsk);
3487 spin_unlock_irqrestore(&list->lock, flags);
3489 EXPORT_SYMBOL(skb_queue_head);
3492 * skb_queue_tail - queue a buffer at the list tail
3493 * @list: list to use
3494 * @newsk: buffer to queue
3496 * Queue a buffer at the tail of the list. This function takes the
3497 * list lock and can be used safely with other locking &sk_buff functions
3500 * A buffer cannot be placed on two lists at the same time.
3502 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3504 unsigned long flags;
3506 spin_lock_irqsave(&list->lock, flags);
3507 __skb_queue_tail(list, newsk);
3508 spin_unlock_irqrestore(&list->lock, flags);
3510 EXPORT_SYMBOL(skb_queue_tail);
3513 * skb_unlink - remove a buffer from a list
3514 * @skb: buffer to remove
3515 * @list: list to use
3517 * Remove a packet from a list. The list locks are taken and this
3518 * function is atomic with respect to other list locked calls
3520 * You must know what list the SKB is on.
3522 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3524 unsigned long flags;
3526 spin_lock_irqsave(&list->lock, flags);
3527 __skb_unlink(skb, list);
3528 spin_unlock_irqrestore(&list->lock, flags);
3530 EXPORT_SYMBOL(skb_unlink);
3533 * skb_append - append a buffer
3534 * @old: buffer to insert after
3535 * @newsk: buffer to insert
3536 * @list: list to use
3538 * Place a packet after a given packet in a list. The list locks are taken
3539 * and this function is atomic with respect to other list locked calls.
3540 * A buffer cannot be placed on two lists at the same time.
3542 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3544 unsigned long flags;
3546 spin_lock_irqsave(&list->lock, flags);
3547 __skb_queue_after(list, old, newsk);
3548 spin_unlock_irqrestore(&list->lock, flags);
3550 EXPORT_SYMBOL(skb_append);
3552 static inline void skb_split_inside_header(struct sk_buff *skb,
3553 struct sk_buff* skb1,
3554 const u32 len, const int pos)
3558 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3560 /* And move data appendix as is. */
3561 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3562 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3564 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3565 skb_shinfo(skb)->nr_frags = 0;
3566 skb1->data_len = skb->data_len;
3567 skb1->len += skb1->data_len;
3570 skb_set_tail_pointer(skb, len);
3573 static inline void skb_split_no_header(struct sk_buff *skb,
3574 struct sk_buff* skb1,
3575 const u32 len, int pos)
3578 const int nfrags = skb_shinfo(skb)->nr_frags;
3580 skb_shinfo(skb)->nr_frags = 0;
3581 skb1->len = skb1->data_len = skb->len - len;
3583 skb->data_len = len - pos;
3585 for (i = 0; i < nfrags; i++) {
3586 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3588 if (pos + size > len) {
3589 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3593 * We have two variants in this case:
3594 * 1. Move all the frag to the second
3595 * part, if it is possible. F.e.
3596 * this approach is mandatory for TUX,
3597 * where splitting is expensive.
3598 * 2. Split is accurately. We make this.
3600 skb_frag_ref(skb, i);
3601 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3602 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3603 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3604 skb_shinfo(skb)->nr_frags++;
3608 skb_shinfo(skb)->nr_frags++;
3611 skb_shinfo(skb1)->nr_frags = k;
3615 * skb_split - Split fragmented skb to two parts at length len.
3616 * @skb: the buffer to split
3617 * @skb1: the buffer to receive the second part
3618 * @len: new length for skb
3620 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3622 int pos = skb_headlen(skb);
3623 const int zc_flags = SKBFL_SHARED_FRAG | SKBFL_PURE_ZEROCOPY;
3625 skb_zcopy_downgrade_managed(skb);
3627 skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & zc_flags;
3628 skb_zerocopy_clone(skb1, skb, 0);
3629 if (len < pos) /* Split line is inside header. */
3630 skb_split_inside_header(skb, skb1, len, pos);
3631 else /* Second chunk has no header, nothing to copy. */
3632 skb_split_no_header(skb, skb1, len, pos);
3634 EXPORT_SYMBOL(skb_split);
3636 /* Shifting from/to a cloned skb is a no-go.
3638 * Caller cannot keep skb_shinfo related pointers past calling here!
3640 static int skb_prepare_for_shift(struct sk_buff *skb)
3642 return skb_unclone_keeptruesize(skb, GFP_ATOMIC);
3646 * skb_shift - Shifts paged data partially from skb to another
3647 * @tgt: buffer into which tail data gets added
3648 * @skb: buffer from which the paged data comes from
3649 * @shiftlen: shift up to this many bytes
3651 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3652 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3653 * It's up to caller to free skb if everything was shifted.
3655 * If @tgt runs out of frags, the whole operation is aborted.
3657 * Skb cannot include anything else but paged data while tgt is allowed
3658 * to have non-paged data as well.
3660 * TODO: full sized shift could be optimized but that would need
3661 * specialized skb free'er to handle frags without up-to-date nr_frags.
3663 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3665 int from, to, merge, todo;
3666 skb_frag_t *fragfrom, *fragto;
3668 BUG_ON(shiftlen > skb->len);
3670 if (skb_headlen(skb))
3672 if (skb_zcopy(tgt) || skb_zcopy(skb))
3677 to = skb_shinfo(tgt)->nr_frags;
3678 fragfrom = &skb_shinfo(skb)->frags[from];
3680 /* Actual merge is delayed until the point when we know we can
3681 * commit all, so that we don't have to undo partial changes
3684 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3685 skb_frag_off(fragfrom))) {
3690 todo -= skb_frag_size(fragfrom);
3692 if (skb_prepare_for_shift(skb) ||
3693 skb_prepare_for_shift(tgt))
3696 /* All previous frag pointers might be stale! */
3697 fragfrom = &skb_shinfo(skb)->frags[from];
3698 fragto = &skb_shinfo(tgt)->frags[merge];
3700 skb_frag_size_add(fragto, shiftlen);
3701 skb_frag_size_sub(fragfrom, shiftlen);
3702 skb_frag_off_add(fragfrom, shiftlen);
3710 /* Skip full, not-fitting skb to avoid expensive operations */
3711 if ((shiftlen == skb->len) &&
3712 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3715 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3718 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3719 if (to == MAX_SKB_FRAGS)
3722 fragfrom = &skb_shinfo(skb)->frags[from];
3723 fragto = &skb_shinfo(tgt)->frags[to];
3725 if (todo >= skb_frag_size(fragfrom)) {
3726 *fragto = *fragfrom;
3727 todo -= skb_frag_size(fragfrom);
3732 __skb_frag_ref(fragfrom);
3733 skb_frag_page_copy(fragto, fragfrom);
3734 skb_frag_off_copy(fragto, fragfrom);
3735 skb_frag_size_set(fragto, todo);
3737 skb_frag_off_add(fragfrom, todo);
3738 skb_frag_size_sub(fragfrom, todo);
3746 /* Ready to "commit" this state change to tgt */
3747 skb_shinfo(tgt)->nr_frags = to;
3750 fragfrom = &skb_shinfo(skb)->frags[0];
3751 fragto = &skb_shinfo(tgt)->frags[merge];
3753 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3754 __skb_frag_unref(fragfrom, skb->pp_recycle);
3757 /* Reposition in the original skb */
3759 while (from < skb_shinfo(skb)->nr_frags)
3760 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3761 skb_shinfo(skb)->nr_frags = to;
3763 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3766 /* Most likely the tgt won't ever need its checksum anymore, skb on
3767 * the other hand might need it if it needs to be resent
3769 tgt->ip_summed = CHECKSUM_PARTIAL;
3770 skb->ip_summed = CHECKSUM_PARTIAL;
3772 skb_len_add(skb, -shiftlen);
3773 skb_len_add(tgt, shiftlen);
3779 * skb_prepare_seq_read - Prepare a sequential read of skb data
3780 * @skb: the buffer to read
3781 * @from: lower offset of data to be read
3782 * @to: upper offset of data to be read
3783 * @st: state variable
3785 * Initializes the specified state variable. Must be called before
3786 * invoking skb_seq_read() for the first time.
3788 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3789 unsigned int to, struct skb_seq_state *st)
3791 st->lower_offset = from;
3792 st->upper_offset = to;
3793 st->root_skb = st->cur_skb = skb;
3794 st->frag_idx = st->stepped_offset = 0;
3795 st->frag_data = NULL;
3798 EXPORT_SYMBOL(skb_prepare_seq_read);
3801 * skb_seq_read - Sequentially read skb data
3802 * @consumed: number of bytes consumed by the caller so far
3803 * @data: destination pointer for data to be returned
3804 * @st: state variable
3806 * Reads a block of skb data at @consumed relative to the
3807 * lower offset specified to skb_prepare_seq_read(). Assigns
3808 * the head of the data block to @data and returns the length
3809 * of the block or 0 if the end of the skb data or the upper
3810 * offset has been reached.
3812 * The caller is not required to consume all of the data
3813 * returned, i.e. @consumed is typically set to the number
3814 * of bytes already consumed and the next call to
3815 * skb_seq_read() will return the remaining part of the block.
3817 * Note 1: The size of each block of data returned can be arbitrary,
3818 * this limitation is the cost for zerocopy sequential
3819 * reads of potentially non linear data.
3821 * Note 2: Fragment lists within fragments are not implemented
3822 * at the moment, state->root_skb could be replaced with
3823 * a stack for this purpose.
3825 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3826 struct skb_seq_state *st)
3828 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3831 if (unlikely(abs_offset >= st->upper_offset)) {
3832 if (st->frag_data) {
3833 kunmap_atomic(st->frag_data);
3834 st->frag_data = NULL;
3840 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3842 if (abs_offset < block_limit && !st->frag_data) {
3843 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3844 return block_limit - abs_offset;
3847 if (st->frag_idx == 0 && !st->frag_data)
3848 st->stepped_offset += skb_headlen(st->cur_skb);
3850 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3851 unsigned int pg_idx, pg_off, pg_sz;
3853 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3856 pg_off = skb_frag_off(frag);
3857 pg_sz = skb_frag_size(frag);
3859 if (skb_frag_must_loop(skb_frag_page(frag))) {
3860 pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT;
3861 pg_off = offset_in_page(pg_off + st->frag_off);
3862 pg_sz = min_t(unsigned int, pg_sz - st->frag_off,
3863 PAGE_SIZE - pg_off);
3866 block_limit = pg_sz + st->stepped_offset;
3867 if (abs_offset < block_limit) {
3869 st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx);
3871 *data = (u8 *)st->frag_data + pg_off +
3872 (abs_offset - st->stepped_offset);
3874 return block_limit - abs_offset;
3877 if (st->frag_data) {
3878 kunmap_atomic(st->frag_data);
3879 st->frag_data = NULL;
3882 st->stepped_offset += pg_sz;
3883 st->frag_off += pg_sz;
3884 if (st->frag_off == skb_frag_size(frag)) {
3890 if (st->frag_data) {
3891 kunmap_atomic(st->frag_data);
3892 st->frag_data = NULL;
3895 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3896 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3899 } else if (st->cur_skb->next) {
3900 st->cur_skb = st->cur_skb->next;
3907 EXPORT_SYMBOL(skb_seq_read);
3910 * skb_abort_seq_read - Abort a sequential read of skb data
3911 * @st: state variable
3913 * Must be called if skb_seq_read() was not called until it
3916 void skb_abort_seq_read(struct skb_seq_state *st)
3919 kunmap_atomic(st->frag_data);
3921 EXPORT_SYMBOL(skb_abort_seq_read);
3923 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3925 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3926 struct ts_config *conf,
3927 struct ts_state *state)
3929 return skb_seq_read(offset, text, TS_SKB_CB(state));
3932 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3934 skb_abort_seq_read(TS_SKB_CB(state));
3938 * skb_find_text - Find a text pattern in skb data
3939 * @skb: the buffer to look in
3940 * @from: search offset
3942 * @config: textsearch configuration
3944 * Finds a pattern in the skb data according to the specified
3945 * textsearch configuration. Use textsearch_next() to retrieve
3946 * subsequent occurrences of the pattern. Returns the offset
3947 * to the first occurrence or UINT_MAX if no match was found.
3949 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3950 unsigned int to, struct ts_config *config)
3952 struct ts_state state;
3955 BUILD_BUG_ON(sizeof(struct skb_seq_state) > sizeof(state.cb));
3957 config->get_next_block = skb_ts_get_next_block;
3958 config->finish = skb_ts_finish;
3960 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3962 ret = textsearch_find(config, &state);
3963 return (ret <= to - from ? ret : UINT_MAX);
3965 EXPORT_SYMBOL(skb_find_text);
3967 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3968 int offset, size_t size)
3970 int i = skb_shinfo(skb)->nr_frags;
3972 if (skb_can_coalesce(skb, i, page, offset)) {
3973 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3974 } else if (i < MAX_SKB_FRAGS) {
3975 skb_zcopy_downgrade_managed(skb);
3977 skb_fill_page_desc_noacc(skb, i, page, offset, size);
3984 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3987 * skb_pull_rcsum - pull skb and update receive checksum
3988 * @skb: buffer to update
3989 * @len: length of data pulled
3991 * This function performs an skb_pull on the packet and updates
3992 * the CHECKSUM_COMPLETE checksum. It should be used on
3993 * receive path processing instead of skb_pull unless you know
3994 * that the checksum difference is zero (e.g., a valid IP header)
3995 * or you are setting ip_summed to CHECKSUM_NONE.
3997 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3999 unsigned char *data = skb->data;
4001 BUG_ON(len > skb->len);
4002 __skb_pull(skb, len);
4003 skb_postpull_rcsum(skb, data, len);
4006 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
4008 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
4010 skb_frag_t head_frag;
4013 page = virt_to_head_page(frag_skb->head);
4014 __skb_frag_set_page(&head_frag, page);
4015 skb_frag_off_set(&head_frag, frag_skb->data -
4016 (unsigned char *)page_address(page));
4017 skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
4021 struct sk_buff *skb_segment_list(struct sk_buff *skb,
4022 netdev_features_t features,
4023 unsigned int offset)
4025 struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
4026 unsigned int tnl_hlen = skb_tnl_header_len(skb);
4027 unsigned int delta_truesize = 0;
4028 unsigned int delta_len = 0;
4029 struct sk_buff *tail = NULL;
4030 struct sk_buff *nskb, *tmp;
4033 skb_push(skb, -skb_network_offset(skb) + offset);
4035 skb_shinfo(skb)->frag_list = NULL;
4039 list_skb = list_skb->next;
4042 delta_truesize += nskb->truesize;
4043 if (skb_shared(nskb)) {
4044 tmp = skb_clone(nskb, GFP_ATOMIC);
4048 err = skb_unclone(nskb, GFP_ATOMIC);
4059 if (unlikely(err)) {
4060 nskb->next = list_skb;
4066 delta_len += nskb->len;
4068 skb_push(nskb, -skb_network_offset(nskb) + offset);
4070 skb_release_head_state(nskb);
4071 len_diff = skb_network_header_len(nskb) - skb_network_header_len(skb);
4072 __copy_skb_header(nskb, skb);
4074 skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
4075 nskb->transport_header += len_diff;
4076 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
4077 nskb->data - tnl_hlen,
4080 if (skb_needs_linearize(nskb, features) &&
4081 __skb_linearize(nskb))
4085 skb->truesize = skb->truesize - delta_truesize;
4086 skb->data_len = skb->data_len - delta_len;
4087 skb->len = skb->len - delta_len;
4093 if (skb_needs_linearize(skb, features) &&
4094 __skb_linearize(skb))
4102 kfree_skb_list(skb->next);
4104 return ERR_PTR(-ENOMEM);
4106 EXPORT_SYMBOL_GPL(skb_segment_list);
4109 * skb_segment - Perform protocol segmentation on skb.
4110 * @head_skb: buffer to segment
4111 * @features: features for the output path (see dev->features)
4113 * This function performs segmentation on the given skb. It returns
4114 * a pointer to the first in a list of new skbs for the segments.
4115 * In case of error it returns ERR_PTR(err).
4117 struct sk_buff *skb_segment(struct sk_buff *head_skb,
4118 netdev_features_t features)
4120 struct sk_buff *segs = NULL;
4121 struct sk_buff *tail = NULL;
4122 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
4123 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
4124 unsigned int mss = skb_shinfo(head_skb)->gso_size;
4125 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
4126 struct sk_buff *frag_skb = head_skb;
4127 unsigned int offset = doffset;
4128 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
4129 unsigned int partial_segs = 0;
4130 unsigned int headroom;
4131 unsigned int len = head_skb->len;
4134 int nfrags = skb_shinfo(head_skb)->nr_frags;
4139 if ((skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY) &&
4140 mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) {
4141 struct sk_buff *check_skb;
4143 for (check_skb = list_skb; check_skb; check_skb = check_skb->next) {
4144 if (skb_headlen(check_skb) && !check_skb->head_frag) {
4145 /* gso_size is untrusted, and we have a frag_list with
4146 * a linear non head_frag item.
4148 * If head_skb's headlen does not fit requested gso_size,
4149 * it means that the frag_list members do NOT terminate
4150 * on exact gso_size boundaries. Hence we cannot perform
4151 * skb_frag_t page sharing. Therefore we must fallback to
4152 * copying the frag_list skbs; we do so by disabling SG.
4154 features &= ~NETIF_F_SG;
4160 __skb_push(head_skb, doffset);
4161 proto = skb_network_protocol(head_skb, NULL);
4162 if (unlikely(!proto))
4163 return ERR_PTR(-EINVAL);
4165 sg = !!(features & NETIF_F_SG);
4166 csum = !!can_checksum_protocol(features, proto);
4168 if (sg && csum && (mss != GSO_BY_FRAGS)) {
4169 if (!(features & NETIF_F_GSO_PARTIAL)) {
4170 struct sk_buff *iter;
4171 unsigned int frag_len;
4174 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
4177 /* If we get here then all the required
4178 * GSO features except frag_list are supported.
4179 * Try to split the SKB to multiple GSO SKBs
4180 * with no frag_list.
4181 * Currently we can do that only when the buffers don't
4182 * have a linear part and all the buffers except
4183 * the last are of the same length.
4185 frag_len = list_skb->len;
4186 skb_walk_frags(head_skb, iter) {
4187 if (frag_len != iter->len && iter->next)
4189 if (skb_headlen(iter) && !iter->head_frag)
4195 if (len != frag_len)
4199 /* GSO partial only requires that we trim off any excess that
4200 * doesn't fit into an MSS sized block, so take care of that
4203 partial_segs = len / mss;
4204 if (partial_segs > 1)
4205 mss *= partial_segs;
4211 headroom = skb_headroom(head_skb);
4212 pos = skb_headlen(head_skb);
4215 struct sk_buff *nskb;
4216 skb_frag_t *nskb_frag;
4220 if (unlikely(mss == GSO_BY_FRAGS)) {
4221 len = list_skb->len;
4223 len = head_skb->len - offset;
4228 hsize = skb_headlen(head_skb) - offset;
4230 if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) &&
4231 (skb_headlen(list_skb) == len || sg)) {
4232 BUG_ON(skb_headlen(list_skb) > len);
4235 nfrags = skb_shinfo(list_skb)->nr_frags;
4236 frag = skb_shinfo(list_skb)->frags;
4237 frag_skb = list_skb;
4238 pos += skb_headlen(list_skb);
4240 while (pos < offset + len) {
4241 BUG_ON(i >= nfrags);
4243 size = skb_frag_size(frag);
4244 if (pos + size > offset + len)
4252 nskb = skb_clone(list_skb, GFP_ATOMIC);
4253 list_skb = list_skb->next;
4255 if (unlikely(!nskb))
4258 if (unlikely(pskb_trim(nskb, len))) {
4263 hsize = skb_end_offset(nskb);
4264 if (skb_cow_head(nskb, doffset + headroom)) {
4269 nskb->truesize += skb_end_offset(nskb) - hsize;
4270 skb_release_head_state(nskb);
4271 __skb_push(nskb, doffset);
4275 if (hsize > len || !sg)
4278 nskb = __alloc_skb(hsize + doffset + headroom,
4279 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
4282 if (unlikely(!nskb))
4285 skb_reserve(nskb, headroom);
4286 __skb_put(nskb, doffset);
4295 __copy_skb_header(nskb, head_skb);
4297 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
4298 skb_reset_mac_len(nskb);
4300 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
4301 nskb->data - tnl_hlen,
4302 doffset + tnl_hlen);
4304 if (nskb->len == len + doffset)
4305 goto perform_csum_check;
4309 if (!nskb->remcsum_offload)
4310 nskb->ip_summed = CHECKSUM_NONE;
4311 SKB_GSO_CB(nskb)->csum =
4312 skb_copy_and_csum_bits(head_skb, offset,
4316 SKB_GSO_CB(nskb)->csum_start =
4317 skb_headroom(nskb) + doffset;
4319 if (skb_copy_bits(head_skb, offset, skb_put(nskb, len), len))
4325 nskb_frag = skb_shinfo(nskb)->frags;
4327 skb_copy_from_linear_data_offset(head_skb, offset,
4328 skb_put(nskb, hsize), hsize);
4330 skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags &
4333 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4334 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
4337 while (pos < offset + len) {
4340 nfrags = skb_shinfo(list_skb)->nr_frags;
4341 frag = skb_shinfo(list_skb)->frags;
4342 frag_skb = list_skb;
4343 if (!skb_headlen(list_skb)) {
4346 BUG_ON(!list_skb->head_frag);
4348 /* to make room for head_frag. */
4352 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4353 skb_zerocopy_clone(nskb, frag_skb,
4357 list_skb = list_skb->next;
4360 if (unlikely(skb_shinfo(nskb)->nr_frags >=
4362 net_warn_ratelimited(
4363 "skb_segment: too many frags: %u %u\n",
4369 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
4370 __skb_frag_ref(nskb_frag);
4371 size = skb_frag_size(nskb_frag);
4374 skb_frag_off_add(nskb_frag, offset - pos);
4375 skb_frag_size_sub(nskb_frag, offset - pos);
4378 skb_shinfo(nskb)->nr_frags++;
4380 if (pos + size <= offset + len) {
4385 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4393 nskb->data_len = len - hsize;
4394 nskb->len += nskb->data_len;
4395 nskb->truesize += nskb->data_len;
4399 if (skb_has_shared_frag(nskb) &&
4400 __skb_linearize(nskb))
4403 if (!nskb->remcsum_offload)
4404 nskb->ip_summed = CHECKSUM_NONE;
4405 SKB_GSO_CB(nskb)->csum =
4406 skb_checksum(nskb, doffset,
4407 nskb->len - doffset, 0);
4408 SKB_GSO_CB(nskb)->csum_start =
4409 skb_headroom(nskb) + doffset;
4411 } while ((offset += len) < head_skb->len);
4413 /* Some callers want to get the end of the list.
4414 * Put it in segs->prev to avoid walking the list.
4415 * (see validate_xmit_skb_list() for example)
4420 struct sk_buff *iter;
4421 int type = skb_shinfo(head_skb)->gso_type;
4422 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
4424 /* Update type to add partial and then remove dodgy if set */
4425 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
4426 type &= ~SKB_GSO_DODGY;
4428 /* Update GSO info and prepare to start updating headers on
4429 * our way back down the stack of protocols.
4431 for (iter = segs; iter; iter = iter->next) {
4432 skb_shinfo(iter)->gso_size = gso_size;
4433 skb_shinfo(iter)->gso_segs = partial_segs;
4434 skb_shinfo(iter)->gso_type = type;
4435 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4438 if (tail->len - doffset <= gso_size)
4439 skb_shinfo(tail)->gso_size = 0;
4440 else if (tail != segs)
4441 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4444 /* Following permits correct backpressure, for protocols
4445 * using skb_set_owner_w().
4446 * Idea is to tranfert ownership from head_skb to last segment.
4448 if (head_skb->destructor == sock_wfree) {
4449 swap(tail->truesize, head_skb->truesize);
4450 swap(tail->destructor, head_skb->destructor);
4451 swap(tail->sk, head_skb->sk);
4456 kfree_skb_list(segs);
4457 return ERR_PTR(err);
4459 EXPORT_SYMBOL_GPL(skb_segment);
4461 #ifdef CONFIG_SKB_EXTENSIONS
4462 #define SKB_EXT_ALIGN_VALUE 8
4463 #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4465 static const u8 skb_ext_type_len[] = {
4466 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4467 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4470 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4472 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4473 [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4475 #if IS_ENABLED(CONFIG_MPTCP)
4476 [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
4478 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
4479 [SKB_EXT_MCTP] = SKB_EXT_CHUNKSIZEOF(struct mctp_flow),
4483 static __always_inline unsigned int skb_ext_total_length(void)
4485 return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4486 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4487 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4490 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4492 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4493 skb_ext_type_len[TC_SKB_EXT] +
4495 #if IS_ENABLED(CONFIG_MPTCP)
4496 skb_ext_type_len[SKB_EXT_MPTCP] +
4498 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
4499 skb_ext_type_len[SKB_EXT_MCTP] +
4504 static void skb_extensions_init(void)
4506 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4507 BUILD_BUG_ON(skb_ext_total_length() > 255);
4509 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4510 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4512 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4516 static void skb_extensions_init(void) {}
4519 void __init skb_init(void)
4521 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4522 sizeof(struct sk_buff),
4524 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4525 offsetof(struct sk_buff, cb),
4526 sizeof_field(struct sk_buff, cb),
4528 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4529 sizeof(struct sk_buff_fclones),
4531 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4533 skb_extensions_init();
4537 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4538 unsigned int recursion_level)
4540 int start = skb_headlen(skb);
4541 int i, copy = start - offset;
4542 struct sk_buff *frag_iter;
4545 if (unlikely(recursion_level >= 24))
4551 sg_set_buf(sg, skb->data + offset, copy);
4553 if ((len -= copy) == 0)
4558 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4561 WARN_ON(start > offset + len);
4563 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4564 if ((copy = end - offset) > 0) {
4565 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4566 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4571 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4572 skb_frag_off(frag) + offset - start);
4581 skb_walk_frags(skb, frag_iter) {
4584 WARN_ON(start > offset + len);
4586 end = start + frag_iter->len;
4587 if ((copy = end - offset) > 0) {
4588 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4593 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4594 copy, recursion_level + 1);
4595 if (unlikely(ret < 0))
4598 if ((len -= copy) == 0)
4609 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4610 * @skb: Socket buffer containing the buffers to be mapped
4611 * @sg: The scatter-gather list to map into
4612 * @offset: The offset into the buffer's contents to start mapping
4613 * @len: Length of buffer space to be mapped
4615 * Fill the specified scatter-gather list with mappings/pointers into a
4616 * region of the buffer space attached to a socket buffer. Returns either
4617 * the number of scatterlist items used, or -EMSGSIZE if the contents
4620 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4622 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4627 sg_mark_end(&sg[nsg - 1]);
4631 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4633 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4634 * sglist without mark the sg which contain last skb data as the end.
4635 * So the caller can mannipulate sg list as will when padding new data after
4636 * the first call without calling sg_unmark_end to expend sg list.
4638 * Scenario to use skb_to_sgvec_nomark:
4640 * 2. skb_to_sgvec_nomark(payload1)
4641 * 3. skb_to_sgvec_nomark(payload2)
4643 * This is equivalent to:
4645 * 2. skb_to_sgvec(payload1)
4647 * 4. skb_to_sgvec(payload2)
4649 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4650 * is more preferable.
4652 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4653 int offset, int len)
4655 return __skb_to_sgvec(skb, sg, offset, len, 0);
4657 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4662 * skb_cow_data - Check that a socket buffer's data buffers are writable
4663 * @skb: The socket buffer to check.
4664 * @tailbits: Amount of trailing space to be added
4665 * @trailer: Returned pointer to the skb where the @tailbits space begins
4667 * Make sure that the data buffers attached to a socket buffer are
4668 * writable. If they are not, private copies are made of the data buffers
4669 * and the socket buffer is set to use these instead.
4671 * If @tailbits is given, make sure that there is space to write @tailbits
4672 * bytes of data beyond current end of socket buffer. @trailer will be
4673 * set to point to the skb in which this space begins.
4675 * The number of scatterlist elements required to completely map the
4676 * COW'd and extended socket buffer will be returned.
4678 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4682 struct sk_buff *skb1, **skb_p;
4684 /* If skb is cloned or its head is paged, reallocate
4685 * head pulling out all the pages (pages are considered not writable
4686 * at the moment even if they are anonymous).
4688 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4689 !__pskb_pull_tail(skb, __skb_pagelen(skb)))
4692 /* Easy case. Most of packets will go this way. */
4693 if (!skb_has_frag_list(skb)) {
4694 /* A little of trouble, not enough of space for trailer.
4695 * This should not happen, when stack is tuned to generate
4696 * good frames. OK, on miss we reallocate and reserve even more
4697 * space, 128 bytes is fair. */
4699 if (skb_tailroom(skb) < tailbits &&
4700 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4708 /* Misery. We are in troubles, going to mincer fragments... */
4711 skb_p = &skb_shinfo(skb)->frag_list;
4714 while ((skb1 = *skb_p) != NULL) {
4717 /* The fragment is partially pulled by someone,
4718 * this can happen on input. Copy it and everything
4721 if (skb_shared(skb1))
4724 /* If the skb is the last, worry about trailer. */
4726 if (skb1->next == NULL && tailbits) {
4727 if (skb_shinfo(skb1)->nr_frags ||
4728 skb_has_frag_list(skb1) ||
4729 skb_tailroom(skb1) < tailbits)
4730 ntail = tailbits + 128;
4736 skb_shinfo(skb1)->nr_frags ||
4737 skb_has_frag_list(skb1)) {
4738 struct sk_buff *skb2;
4740 /* Fuck, we are miserable poor guys... */
4742 skb2 = skb_copy(skb1, GFP_ATOMIC);
4744 skb2 = skb_copy_expand(skb1,
4748 if (unlikely(skb2 == NULL))
4752 skb_set_owner_w(skb2, skb1->sk);
4754 /* Looking around. Are we still alive?
4755 * OK, link new skb, drop old one */
4757 skb2->next = skb1->next;
4764 skb_p = &skb1->next;
4769 EXPORT_SYMBOL_GPL(skb_cow_data);
4771 static void sock_rmem_free(struct sk_buff *skb)
4773 struct sock *sk = skb->sk;
4775 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4778 static void skb_set_err_queue(struct sk_buff *skb)
4780 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4781 * So, it is safe to (mis)use it to mark skbs on the error queue.
4783 skb->pkt_type = PACKET_OUTGOING;
4784 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4788 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4790 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4792 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4793 (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4798 skb->destructor = sock_rmem_free;
4799 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4800 skb_set_err_queue(skb);
4802 /* before exiting rcu section, make sure dst is refcounted */
4805 skb_queue_tail(&sk->sk_error_queue, skb);
4806 if (!sock_flag(sk, SOCK_DEAD))
4807 sk_error_report(sk);
4810 EXPORT_SYMBOL(sock_queue_err_skb);
4812 static bool is_icmp_err_skb(const struct sk_buff *skb)
4814 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4815 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4818 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4820 struct sk_buff_head *q = &sk->sk_error_queue;
4821 struct sk_buff *skb, *skb_next = NULL;
4822 bool icmp_next = false;
4823 unsigned long flags;
4825 spin_lock_irqsave(&q->lock, flags);
4826 skb = __skb_dequeue(q);
4827 if (skb && (skb_next = skb_peek(q))) {
4828 icmp_next = is_icmp_err_skb(skb_next);
4830 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
4832 spin_unlock_irqrestore(&q->lock, flags);
4834 if (is_icmp_err_skb(skb) && !icmp_next)
4838 sk_error_report(sk);
4842 EXPORT_SYMBOL(sock_dequeue_err_skb);
4845 * skb_clone_sk - create clone of skb, and take reference to socket
4846 * @skb: the skb to clone
4848 * This function creates a clone of a buffer that holds a reference on
4849 * sk_refcnt. Buffers created via this function are meant to be
4850 * returned using sock_queue_err_skb, or free via kfree_skb.
4852 * When passing buffers allocated with this function to sock_queue_err_skb
4853 * it is necessary to wrap the call with sock_hold/sock_put in order to
4854 * prevent the socket from being released prior to being enqueued on
4855 * the sk_error_queue.
4857 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4859 struct sock *sk = skb->sk;
4860 struct sk_buff *clone;
4862 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4865 clone = skb_clone(skb, GFP_ATOMIC);
4872 clone->destructor = sock_efree;
4876 EXPORT_SYMBOL(skb_clone_sk);
4878 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4883 struct sock_exterr_skb *serr;
4886 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4888 serr = SKB_EXT_ERR(skb);
4889 memset(serr, 0, sizeof(*serr));
4890 serr->ee.ee_errno = ENOMSG;
4891 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4892 serr->ee.ee_info = tstype;
4893 serr->opt_stats = opt_stats;
4894 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4895 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4896 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4898 serr->ee.ee_data -= atomic_read(&sk->sk_tskey);
4901 err = sock_queue_err_skb(sk, skb);
4907 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4911 if (likely(READ_ONCE(sysctl_tstamp_allow_data) || tsonly))
4914 read_lock_bh(&sk->sk_callback_lock);
4915 ret = sk->sk_socket && sk->sk_socket->file &&
4916 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4917 read_unlock_bh(&sk->sk_callback_lock);
4921 void skb_complete_tx_timestamp(struct sk_buff *skb,
4922 struct skb_shared_hwtstamps *hwtstamps)
4924 struct sock *sk = skb->sk;
4926 if (!skb_may_tx_timestamp(sk, false))
4929 /* Take a reference to prevent skb_orphan() from freeing the socket,
4930 * but only if the socket refcount is not zero.
4932 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4933 *skb_hwtstamps(skb) = *hwtstamps;
4934 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4942 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4944 void __skb_tstamp_tx(struct sk_buff *orig_skb,
4945 const struct sk_buff *ack_skb,
4946 struct skb_shared_hwtstamps *hwtstamps,
4947 struct sock *sk, int tstype)
4949 struct sk_buff *skb;
4950 bool tsonly, opt_stats = false;
4955 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4956 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4959 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4960 if (!skb_may_tx_timestamp(sk, tsonly))
4965 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4967 skb = tcp_get_timestamping_opt_stats(sk, orig_skb,
4972 skb = alloc_skb(0, GFP_ATOMIC);
4974 skb = skb_clone(orig_skb, GFP_ATOMIC);
4980 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4982 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4986 *skb_hwtstamps(skb) = *hwtstamps;
4988 __net_timestamp(skb);
4990 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4992 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4994 void skb_tstamp_tx(struct sk_buff *orig_skb,
4995 struct skb_shared_hwtstamps *hwtstamps)
4997 return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk,
5000 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
5002 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
5004 struct sock *sk = skb->sk;
5005 struct sock_exterr_skb *serr;
5008 skb->wifi_acked_valid = 1;
5009 skb->wifi_acked = acked;
5011 serr = SKB_EXT_ERR(skb);
5012 memset(serr, 0, sizeof(*serr));
5013 serr->ee.ee_errno = ENOMSG;
5014 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
5016 /* Take a reference to prevent skb_orphan() from freeing the socket,
5017 * but only if the socket refcount is not zero.
5019 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
5020 err = sock_queue_err_skb(sk, skb);
5026 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
5029 * skb_partial_csum_set - set up and verify partial csum values for packet
5030 * @skb: the skb to set
5031 * @start: the number of bytes after skb->data to start checksumming.
5032 * @off: the offset from start to place the checksum.
5034 * For untrusted partially-checksummed packets, we need to make sure the values
5035 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
5037 * This function checks and sets those values and skb->ip_summed: if this
5038 * returns false you should drop the packet.
5040 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
5042 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
5043 u32 csum_start = skb_headroom(skb) + (u32)start;
5045 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
5046 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
5047 start, off, skb_headroom(skb), skb_headlen(skb));
5050 skb->ip_summed = CHECKSUM_PARTIAL;
5051 skb->csum_start = csum_start;
5052 skb->csum_offset = off;
5053 skb_set_transport_header(skb, start);
5056 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
5058 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
5061 if (skb_headlen(skb) >= len)
5064 /* If we need to pullup then pullup to the max, so we
5065 * won't need to do it again.
5070 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
5073 if (skb_headlen(skb) < len)
5079 #define MAX_TCP_HDR_LEN (15 * 4)
5081 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
5082 typeof(IPPROTO_IP) proto,
5089 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
5090 off + MAX_TCP_HDR_LEN);
5091 if (!err && !skb_partial_csum_set(skb, off,
5092 offsetof(struct tcphdr,
5095 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
5098 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
5099 off + sizeof(struct udphdr));
5100 if (!err && !skb_partial_csum_set(skb, off,
5101 offsetof(struct udphdr,
5104 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
5107 return ERR_PTR(-EPROTO);
5110 /* This value should be large enough to cover a tagged ethernet header plus
5111 * maximally sized IP and TCP or UDP headers.
5113 #define MAX_IP_HDR_LEN 128
5115 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
5124 err = skb_maybe_pull_tail(skb,
5125 sizeof(struct iphdr),
5130 if (ip_is_fragment(ip_hdr(skb)))
5133 off = ip_hdrlen(skb);
5140 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
5142 return PTR_ERR(csum);
5145 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
5148 ip_hdr(skb)->protocol, 0);
5155 /* This value should be large enough to cover a tagged ethernet header plus
5156 * an IPv6 header, all options, and a maximal TCP or UDP header.
5158 #define MAX_IPV6_HDR_LEN 256
5160 #define OPT_HDR(type, skb, off) \
5161 (type *)(skb_network_header(skb) + (off))
5163 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
5176 off = sizeof(struct ipv6hdr);
5178 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
5182 nexthdr = ipv6_hdr(skb)->nexthdr;
5184 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
5185 while (off <= len && !done) {
5187 case IPPROTO_DSTOPTS:
5188 case IPPROTO_HOPOPTS:
5189 case IPPROTO_ROUTING: {
5190 struct ipv6_opt_hdr *hp;
5192 err = skb_maybe_pull_tail(skb,
5194 sizeof(struct ipv6_opt_hdr),
5199 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
5200 nexthdr = hp->nexthdr;
5201 off += ipv6_optlen(hp);
5205 struct ip_auth_hdr *hp;
5207 err = skb_maybe_pull_tail(skb,
5209 sizeof(struct ip_auth_hdr),
5214 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
5215 nexthdr = hp->nexthdr;
5216 off += ipv6_authlen(hp);
5219 case IPPROTO_FRAGMENT: {
5220 struct frag_hdr *hp;
5222 err = skb_maybe_pull_tail(skb,
5224 sizeof(struct frag_hdr),
5229 hp = OPT_HDR(struct frag_hdr, skb, off);
5231 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
5234 nexthdr = hp->nexthdr;
5235 off += sizeof(struct frag_hdr);
5246 if (!done || fragment)
5249 csum = skb_checksum_setup_ip(skb, nexthdr, off);
5251 return PTR_ERR(csum);
5254 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5255 &ipv6_hdr(skb)->daddr,
5256 skb->len - off, nexthdr, 0);
5264 * skb_checksum_setup - set up partial checksum offset
5265 * @skb: the skb to set up
5266 * @recalculate: if true the pseudo-header checksum will be recalculated
5268 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
5272 switch (skb->protocol) {
5273 case htons(ETH_P_IP):
5274 err = skb_checksum_setup_ipv4(skb, recalculate);
5277 case htons(ETH_P_IPV6):
5278 err = skb_checksum_setup_ipv6(skb, recalculate);
5288 EXPORT_SYMBOL(skb_checksum_setup);
5291 * skb_checksum_maybe_trim - maybe trims the given skb
5292 * @skb: the skb to check
5293 * @transport_len: the data length beyond the network header
5295 * Checks whether the given skb has data beyond the given transport length.
5296 * If so, returns a cloned skb trimmed to this transport length.
5297 * Otherwise returns the provided skb. Returns NULL in error cases
5298 * (e.g. transport_len exceeds skb length or out-of-memory).
5300 * Caller needs to set the skb transport header and free any returned skb if it
5301 * differs from the provided skb.
5303 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5304 unsigned int transport_len)
5306 struct sk_buff *skb_chk;
5307 unsigned int len = skb_transport_offset(skb) + transport_len;
5312 else if (skb->len == len)
5315 skb_chk = skb_clone(skb, GFP_ATOMIC);
5319 ret = pskb_trim_rcsum(skb_chk, len);
5329 * skb_checksum_trimmed - validate checksum of an skb
5330 * @skb: the skb to check
5331 * @transport_len: the data length beyond the network header
5332 * @skb_chkf: checksum function to use
5334 * Applies the given checksum function skb_chkf to the provided skb.
5335 * Returns a checked and maybe trimmed skb. Returns NULL on error.
5337 * If the skb has data beyond the given transport length, then a
5338 * trimmed & cloned skb is checked and returned.
5340 * Caller needs to set the skb transport header and free any returned skb if it
5341 * differs from the provided skb.
5343 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5344 unsigned int transport_len,
5345 __sum16(*skb_chkf)(struct sk_buff *skb))
5347 struct sk_buff *skb_chk;
5348 unsigned int offset = skb_transport_offset(skb);
5351 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5355 if (!pskb_may_pull(skb_chk, offset))
5358 skb_pull_rcsum(skb_chk, offset);
5359 ret = skb_chkf(skb_chk);
5360 skb_push_rcsum(skb_chk, offset);
5368 if (skb_chk && skb_chk != skb)
5374 EXPORT_SYMBOL(skb_checksum_trimmed);
5376 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5378 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5381 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5383 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5386 skb_release_head_state(skb);
5387 kmem_cache_free(skbuff_head_cache, skb);
5392 EXPORT_SYMBOL(kfree_skb_partial);
5395 * skb_try_coalesce - try to merge skb to prior one
5397 * @from: buffer to add
5398 * @fragstolen: pointer to boolean
5399 * @delta_truesize: how much more was allocated than was requested
5401 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5402 bool *fragstolen, int *delta_truesize)
5404 struct skb_shared_info *to_shinfo, *from_shinfo;
5405 int i, delta, len = from->len;
5407 *fragstolen = false;
5412 /* In general, avoid mixing slab allocated and page_pool allocated
5413 * pages within the same SKB. However when @to is not pp_recycle and
5414 * @from is cloned, we can transition frag pages from page_pool to
5415 * reference counted.
5417 * On the other hand, don't allow coalescing two pp_recycle SKBs if
5418 * @from is cloned, in case the SKB is using page_pool fragment
5419 * references (PP_FLAG_PAGE_FRAG). Since we only take full page
5420 * references for cloned SKBs at the moment that would result in
5421 * inconsistent reference counts.
5423 if (to->pp_recycle != (from->pp_recycle && !skb_cloned(from)))
5426 if (len <= skb_tailroom(to)) {
5428 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5429 *delta_truesize = 0;
5433 to_shinfo = skb_shinfo(to);
5434 from_shinfo = skb_shinfo(from);
5435 if (to_shinfo->frag_list || from_shinfo->frag_list)
5437 if (skb_zcopy(to) || skb_zcopy(from))
5440 if (skb_headlen(from) != 0) {
5442 unsigned int offset;
5444 if (to_shinfo->nr_frags +
5445 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5448 if (skb_head_is_locked(from))
5451 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5453 page = virt_to_head_page(from->head);
5454 offset = from->data - (unsigned char *)page_address(page);
5456 skb_fill_page_desc(to, to_shinfo->nr_frags,
5457 page, offset, skb_headlen(from));
5460 if (to_shinfo->nr_frags +
5461 from_shinfo->nr_frags > MAX_SKB_FRAGS)
5464 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5467 WARN_ON_ONCE(delta < len);
5469 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5471 from_shinfo->nr_frags * sizeof(skb_frag_t));
5472 to_shinfo->nr_frags += from_shinfo->nr_frags;
5474 if (!skb_cloned(from))
5475 from_shinfo->nr_frags = 0;
5477 /* if the skb is not cloned this does nothing
5478 * since we set nr_frags to 0.
5480 for (i = 0; i < from_shinfo->nr_frags; i++)
5481 __skb_frag_ref(&from_shinfo->frags[i]);
5483 to->truesize += delta;
5485 to->data_len += len;
5487 *delta_truesize = delta;
5490 EXPORT_SYMBOL(skb_try_coalesce);
5493 * skb_scrub_packet - scrub an skb
5495 * @skb: buffer to clean
5496 * @xnet: packet is crossing netns
5498 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5499 * into/from a tunnel. Some information have to be cleared during these
5501 * skb_scrub_packet can also be used to clean a skb before injecting it in
5502 * another namespace (@xnet == true). We have to clear all information in the
5503 * skb that could impact namespace isolation.
5505 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5507 skb->pkt_type = PACKET_HOST;
5513 nf_reset_trace(skb);
5515 #ifdef CONFIG_NET_SWITCHDEV
5516 skb->offload_fwd_mark = 0;
5517 skb->offload_l3_fwd_mark = 0;
5525 skb_clear_tstamp(skb);
5527 EXPORT_SYMBOL_GPL(skb_scrub_packet);
5530 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5534 * skb_gso_transport_seglen is used to determine the real size of the
5535 * individual segments, including Layer4 headers (TCP/UDP).
5537 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5539 static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5541 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5542 unsigned int thlen = 0;
5544 if (skb->encapsulation) {
5545 thlen = skb_inner_transport_header(skb) -
5546 skb_transport_header(skb);
5548 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5549 thlen += inner_tcp_hdrlen(skb);
5550 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5551 thlen = tcp_hdrlen(skb);
5552 } else if (unlikely(skb_is_gso_sctp(skb))) {
5553 thlen = sizeof(struct sctphdr);
5554 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5555 thlen = sizeof(struct udphdr);
5557 /* UFO sets gso_size to the size of the fragmentation
5558 * payload, i.e. the size of the L4 (UDP) header is already
5561 return thlen + shinfo->gso_size;
5565 * skb_gso_network_seglen - Return length of individual segments of a gso packet
5569 * skb_gso_network_seglen is used to determine the real size of the
5570 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5572 * The MAC/L2 header is not accounted for.
5574 static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5576 unsigned int hdr_len = skb_transport_header(skb) -
5577 skb_network_header(skb);
5579 return hdr_len + skb_gso_transport_seglen(skb);
5583 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5587 * skb_gso_mac_seglen is used to determine the real size of the
5588 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5589 * headers (TCP/UDP).
5591 static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5593 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5595 return hdr_len + skb_gso_transport_seglen(skb);
5599 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5601 * There are a couple of instances where we have a GSO skb, and we
5602 * want to determine what size it would be after it is segmented.
5604 * We might want to check:
5605 * - L3+L4+payload size (e.g. IP forwarding)
5606 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5608 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5612 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5613 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5615 * @max_len: The maximum permissible length.
5617 * Returns true if the segmented length <= max length.
5619 static inline bool skb_gso_size_check(const struct sk_buff *skb,
5620 unsigned int seg_len,
5621 unsigned int max_len) {
5622 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5623 const struct sk_buff *iter;
5625 if (shinfo->gso_size != GSO_BY_FRAGS)
5626 return seg_len <= max_len;
5628 /* Undo this so we can re-use header sizes */
5629 seg_len -= GSO_BY_FRAGS;
5631 skb_walk_frags(skb, iter) {
5632 if (seg_len + skb_headlen(iter) > max_len)
5640 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5643 * @mtu: MTU to validate against
5645 * skb_gso_validate_network_len validates if a given skb will fit a
5646 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5649 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5651 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5653 EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5656 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5659 * @len: length to validate against
5661 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5662 * length once split, including L2, L3 and L4 headers and the payload.
5664 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5666 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5668 EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5670 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5672 int mac_len, meta_len;
5675 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5680 mac_len = skb->data - skb_mac_header(skb);
5681 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5682 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5683 mac_len - VLAN_HLEN - ETH_TLEN);
5686 meta_len = skb_metadata_len(skb);
5688 meta = skb_metadata_end(skb) - meta_len;
5689 memmove(meta + VLAN_HLEN, meta, meta_len);
5692 skb->mac_header += VLAN_HLEN;
5696 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5698 struct vlan_hdr *vhdr;
5701 if (unlikely(skb_vlan_tag_present(skb))) {
5702 /* vlan_tci is already set-up so leave this for another time */
5706 skb = skb_share_check(skb, GFP_ATOMIC);
5709 /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
5710 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
5713 vhdr = (struct vlan_hdr *)skb->data;
5714 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5715 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5717 skb_pull_rcsum(skb, VLAN_HLEN);
5718 vlan_set_encap_proto(skb, vhdr);
5720 skb = skb_reorder_vlan_header(skb);
5724 skb_reset_network_header(skb);
5725 if (!skb_transport_header_was_set(skb))
5726 skb_reset_transport_header(skb);
5727 skb_reset_mac_len(skb);
5735 EXPORT_SYMBOL(skb_vlan_untag);
5737 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len)
5739 if (!pskb_may_pull(skb, write_len))
5742 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5745 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5747 EXPORT_SYMBOL(skb_ensure_writable);
5749 /* remove VLAN header from packet and update csum accordingly.
5750 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5752 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5754 struct vlan_hdr *vhdr;
5755 int offset = skb->data - skb_mac_header(skb);
5758 if (WARN_ONCE(offset,
5759 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5764 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5768 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5770 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5771 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5773 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5774 __skb_pull(skb, VLAN_HLEN);
5776 vlan_set_encap_proto(skb, vhdr);
5777 skb->mac_header += VLAN_HLEN;
5779 if (skb_network_offset(skb) < ETH_HLEN)
5780 skb_set_network_header(skb, ETH_HLEN);
5782 skb_reset_mac_len(skb);
5786 EXPORT_SYMBOL(__skb_vlan_pop);
5788 /* Pop a vlan tag either from hwaccel or from payload.
5789 * Expects skb->data at mac header.
5791 int skb_vlan_pop(struct sk_buff *skb)
5797 if (likely(skb_vlan_tag_present(skb))) {
5798 __vlan_hwaccel_clear_tag(skb);
5800 if (unlikely(!eth_type_vlan(skb->protocol)))
5803 err = __skb_vlan_pop(skb, &vlan_tci);
5807 /* move next vlan tag to hw accel tag */
5808 if (likely(!eth_type_vlan(skb->protocol)))
5811 vlan_proto = skb->protocol;
5812 err = __skb_vlan_pop(skb, &vlan_tci);
5816 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5819 EXPORT_SYMBOL(skb_vlan_pop);
5821 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5822 * Expects skb->data at mac header.
5824 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5826 if (skb_vlan_tag_present(skb)) {
5827 int offset = skb->data - skb_mac_header(skb);
5830 if (WARN_ONCE(offset,
5831 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5836 err = __vlan_insert_tag(skb, skb->vlan_proto,
5837 skb_vlan_tag_get(skb));
5841 skb->protocol = skb->vlan_proto;
5842 skb->mac_len += VLAN_HLEN;
5844 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5846 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5849 EXPORT_SYMBOL(skb_vlan_push);
5852 * skb_eth_pop() - Drop the Ethernet header at the head of a packet
5854 * @skb: Socket buffer to modify
5856 * Drop the Ethernet header of @skb.
5858 * Expects that skb->data points to the mac header and that no VLAN tags are
5861 * Returns 0 on success, -errno otherwise.
5863 int skb_eth_pop(struct sk_buff *skb)
5865 if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) ||
5866 skb_network_offset(skb) < ETH_HLEN)
5869 skb_pull_rcsum(skb, ETH_HLEN);
5870 skb_reset_mac_header(skb);
5871 skb_reset_mac_len(skb);
5875 EXPORT_SYMBOL(skb_eth_pop);
5878 * skb_eth_push() - Add a new Ethernet header at the head of a packet
5880 * @skb: Socket buffer to modify
5881 * @dst: Destination MAC address of the new header
5882 * @src: Source MAC address of the new header
5884 * Prepend @skb with a new Ethernet header.
5886 * Expects that skb->data points to the mac header, which must be empty.
5888 * Returns 0 on success, -errno otherwise.
5890 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
5891 const unsigned char *src)
5896 if (skb_network_offset(skb) || skb_vlan_tag_present(skb))
5899 err = skb_cow_head(skb, sizeof(*eth));
5903 skb_push(skb, sizeof(*eth));
5904 skb_reset_mac_header(skb);
5905 skb_reset_mac_len(skb);
5908 ether_addr_copy(eth->h_dest, dst);
5909 ether_addr_copy(eth->h_source, src);
5910 eth->h_proto = skb->protocol;
5912 skb_postpush_rcsum(skb, eth, sizeof(*eth));
5916 EXPORT_SYMBOL(skb_eth_push);
5918 /* Update the ethertype of hdr and the skb csum value if required. */
5919 static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5922 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5923 __be16 diff[] = { ~hdr->h_proto, ethertype };
5925 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5928 hdr->h_proto = ethertype;
5932 * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
5936 * @mpls_lse: MPLS label stack entry to push
5937 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
5938 * @mac_len: length of the MAC header
5939 * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
5942 * Expects skb->data at mac header.
5944 * Returns 0 on success, -errno otherwise.
5946 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
5947 int mac_len, bool ethernet)
5949 struct mpls_shim_hdr *lse;
5952 if (unlikely(!eth_p_mpls(mpls_proto)))
5955 /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
5956 if (skb->encapsulation)
5959 err = skb_cow_head(skb, MPLS_HLEN);
5963 if (!skb->inner_protocol) {
5964 skb_set_inner_network_header(skb, skb_network_offset(skb));
5965 skb_set_inner_protocol(skb, skb->protocol);
5968 skb_push(skb, MPLS_HLEN);
5969 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
5971 skb_reset_mac_header(skb);
5972 skb_set_network_header(skb, mac_len);
5973 skb_reset_mac_len(skb);
5975 lse = mpls_hdr(skb);
5976 lse->label_stack_entry = mpls_lse;
5977 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
5979 if (ethernet && mac_len >= ETH_HLEN)
5980 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
5981 skb->protocol = mpls_proto;
5985 EXPORT_SYMBOL_GPL(skb_mpls_push);
5988 * skb_mpls_pop() - pop the outermost MPLS header
5991 * @next_proto: ethertype of header after popped MPLS header
5992 * @mac_len: length of the MAC header
5993 * @ethernet: flag to indicate if the packet is ethernet
5995 * Expects skb->data at mac header.
5997 * Returns 0 on success, -errno otherwise.
5999 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
6004 if (unlikely(!eth_p_mpls(skb->protocol)))
6007 err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
6011 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
6012 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
6015 __skb_pull(skb, MPLS_HLEN);
6016 skb_reset_mac_header(skb);
6017 skb_set_network_header(skb, mac_len);
6019 if (ethernet && mac_len >= ETH_HLEN) {
6022 /* use mpls_hdr() to get ethertype to account for VLANs. */
6023 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
6024 skb_mod_eth_type(skb, hdr, next_proto);
6026 skb->protocol = next_proto;
6030 EXPORT_SYMBOL_GPL(skb_mpls_pop);
6033 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
6036 * @mpls_lse: new MPLS label stack entry to update to
6038 * Expects skb->data at mac header.
6040 * Returns 0 on success, -errno otherwise.
6042 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
6046 if (unlikely(!eth_p_mpls(skb->protocol)))
6049 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
6053 if (skb->ip_summed == CHECKSUM_COMPLETE) {
6054 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
6056 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
6059 mpls_hdr(skb)->label_stack_entry = mpls_lse;
6063 EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
6066 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
6070 * Expects skb->data at mac header.
6072 * Returns 0 on success, -errno otherwise.
6074 int skb_mpls_dec_ttl(struct sk_buff *skb)
6079 if (unlikely(!eth_p_mpls(skb->protocol)))
6082 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
6085 lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
6086 ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
6090 lse &= ~MPLS_LS_TTL_MASK;
6091 lse |= ttl << MPLS_LS_TTL_SHIFT;
6093 return skb_mpls_update_lse(skb, cpu_to_be32(lse));
6095 EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
6098 * alloc_skb_with_frags - allocate skb with page frags
6100 * @header_len: size of linear part
6101 * @data_len: needed length in frags
6102 * @max_page_order: max page order desired.
6103 * @errcode: pointer to error code if any
6104 * @gfp_mask: allocation mask
6106 * This can be used to allocate a paged skb, given a maximal order for frags.
6108 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
6109 unsigned long data_len,
6114 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
6115 unsigned long chunk;
6116 struct sk_buff *skb;
6120 *errcode = -EMSGSIZE;
6121 /* Note this test could be relaxed, if we succeed to allocate
6122 * high order pages...
6124 if (npages > MAX_SKB_FRAGS)
6127 *errcode = -ENOBUFS;
6128 skb = alloc_skb(header_len, gfp_mask);
6132 skb->truesize += npages << PAGE_SHIFT;
6134 for (i = 0; npages > 0; i++) {
6135 int order = max_page_order;
6138 if (npages >= 1 << order) {
6139 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
6145 /* Do not retry other high order allocations */
6151 page = alloc_page(gfp_mask);
6155 chunk = min_t(unsigned long, data_len,
6156 PAGE_SIZE << order);
6157 skb_fill_page_desc(skb, i, page, 0, chunk);
6159 npages -= 1 << order;
6167 EXPORT_SYMBOL(alloc_skb_with_frags);
6169 /* carve out the first off bytes from skb when off < headlen */
6170 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
6171 const int headlen, gfp_t gfp_mask)
6174 int size = skb_end_offset(skb);
6175 int new_hlen = headlen - off;
6178 size = SKB_DATA_ALIGN(size);
6180 if (skb_pfmemalloc(skb))
6181 gfp_mask |= __GFP_MEMALLOC;
6182 data = kmalloc_reserve(size +
6183 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
6184 gfp_mask, NUMA_NO_NODE, NULL);
6188 size = SKB_WITH_OVERHEAD(ksize(data));
6190 /* Copy real data, and all frags */
6191 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
6194 memcpy((struct skb_shared_info *)(data + size),
6196 offsetof(struct skb_shared_info,
6197 frags[skb_shinfo(skb)->nr_frags]));
6198 if (skb_cloned(skb)) {
6199 /* drop the old head gracefully */
6200 if (skb_orphan_frags(skb, gfp_mask)) {
6204 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
6205 skb_frag_ref(skb, i);
6206 if (skb_has_frag_list(skb))
6207 skb_clone_fraglist(skb);
6208 skb_release_data(skb);
6210 /* we can reuse existing recount- all we did was
6219 skb_set_end_offset(skb, size);
6220 skb_set_tail_pointer(skb, skb_headlen(skb));
6221 skb_headers_offset_update(skb, 0);
6225 atomic_set(&skb_shinfo(skb)->dataref, 1);
6230 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
6232 /* carve out the first eat bytes from skb's frag_list. May recurse into
6235 static int pskb_carve_frag_list(struct sk_buff *skb,
6236 struct skb_shared_info *shinfo, int eat,
6239 struct sk_buff *list = shinfo->frag_list;
6240 struct sk_buff *clone = NULL;
6241 struct sk_buff *insp = NULL;
6245 pr_err("Not enough bytes to eat. Want %d\n", eat);
6248 if (list->len <= eat) {
6249 /* Eaten as whole. */
6254 /* Eaten partially. */
6255 if (skb_shared(list)) {
6256 clone = skb_clone(list, gfp_mask);
6262 /* This may be pulled without problems. */
6265 if (pskb_carve(list, eat, gfp_mask) < 0) {
6273 /* Free pulled out fragments. */
6274 while ((list = shinfo->frag_list) != insp) {
6275 shinfo->frag_list = list->next;
6278 /* And insert new clone at head. */
6281 shinfo->frag_list = clone;
6286 /* carve off first len bytes from skb. Split line (off) is in the
6287 * non-linear part of skb
6289 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
6290 int pos, gfp_t gfp_mask)
6293 int size = skb_end_offset(skb);
6295 const int nfrags = skb_shinfo(skb)->nr_frags;
6296 struct skb_shared_info *shinfo;
6298 size = SKB_DATA_ALIGN(size);
6300 if (skb_pfmemalloc(skb))
6301 gfp_mask |= __GFP_MEMALLOC;
6302 data = kmalloc_reserve(size +
6303 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
6304 gfp_mask, NUMA_NO_NODE, NULL);
6308 size = SKB_WITH_OVERHEAD(ksize(data));
6310 memcpy((struct skb_shared_info *)(data + size),
6311 skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0]));
6312 if (skb_orphan_frags(skb, gfp_mask)) {
6316 shinfo = (struct skb_shared_info *)(data + size);
6317 for (i = 0; i < nfrags; i++) {
6318 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
6320 if (pos + fsize > off) {
6321 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
6325 * We have two variants in this case:
6326 * 1. Move all the frag to the second
6327 * part, if it is possible. F.e.
6328 * this approach is mandatory for TUX,
6329 * where splitting is expensive.
6330 * 2. Split is accurately. We make this.
6332 skb_frag_off_add(&shinfo->frags[0], off - pos);
6333 skb_frag_size_sub(&shinfo->frags[0], off - pos);
6335 skb_frag_ref(skb, i);
6340 shinfo->nr_frags = k;
6341 if (skb_has_frag_list(skb))
6342 skb_clone_fraglist(skb);
6344 /* split line is in frag list */
6345 if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
6346 /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
6347 if (skb_has_frag_list(skb))
6348 kfree_skb_list(skb_shinfo(skb)->frag_list);
6352 skb_release_data(skb);
6357 skb_set_end_offset(skb, size);
6358 skb_reset_tail_pointer(skb);
6359 skb_headers_offset_update(skb, 0);
6364 skb->data_len = skb->len;
6365 atomic_set(&skb_shinfo(skb)->dataref, 1);
6369 /* remove len bytes from the beginning of the skb */
6370 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6372 int headlen = skb_headlen(skb);
6375 return pskb_carve_inside_header(skb, len, headlen, gfp);
6377 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6380 /* Extract to_copy bytes starting at off from skb, and return this in
6383 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6384 int to_copy, gfp_t gfp)
6386 struct sk_buff *clone = skb_clone(skb, gfp);
6391 if (pskb_carve(clone, off, gfp) < 0 ||
6392 pskb_trim(clone, to_copy)) {
6398 EXPORT_SYMBOL(pskb_extract);
6401 * skb_condense - try to get rid of fragments/frag_list if possible
6404 * Can be used to save memory before skb is added to a busy queue.
6405 * If packet has bytes in frags and enough tail room in skb->head,
6406 * pull all of them, so that we can free the frags right now and adjust
6409 * We do not reallocate skb->head thus can not fail.
6410 * Caller must re-evaluate skb->truesize if needed.
6412 void skb_condense(struct sk_buff *skb)
6414 if (skb->data_len) {
6415 if (skb->data_len > skb->end - skb->tail ||
6419 /* Nice, we can free page frag(s) right now */
6420 __pskb_pull_tail(skb, skb->data_len);
6422 /* At this point, skb->truesize might be over estimated,
6423 * because skb had a fragment, and fragments do not tell
6425 * When we pulled its content into skb->head, fragment
6426 * was freed, but __pskb_pull_tail() could not possibly
6427 * adjust skb->truesize, not knowing the frag truesize.
6429 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6432 #ifdef CONFIG_SKB_EXTENSIONS
6433 static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6435 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6439 * __skb_ext_alloc - allocate a new skb extensions storage
6441 * @flags: See kmalloc().
6443 * Returns the newly allocated pointer. The pointer can later attached to a
6444 * skb via __skb_ext_set().
6445 * Note: caller must handle the skb_ext as an opaque data.
6447 struct skb_ext *__skb_ext_alloc(gfp_t flags)
6449 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6452 memset(new->offset, 0, sizeof(new->offset));
6453 refcount_set(&new->refcnt, 1);
6459 static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6460 unsigned int old_active)
6462 struct skb_ext *new;
6464 if (refcount_read(&old->refcnt) == 1)
6467 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6471 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6472 refcount_set(&new->refcnt, 1);
6475 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6476 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6479 for (i = 0; i < sp->len; i++)
6480 xfrm_state_hold(sp->xvec[i]);
6488 * __skb_ext_set - attach the specified extension storage to this skb
6491 * @ext: extension storage previously allocated via __skb_ext_alloc()
6493 * Existing extensions, if any, are cleared.
6495 * Returns the pointer to the extension.
6497 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6498 struct skb_ext *ext)
6500 unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6503 newlen = newoff + skb_ext_type_len[id];
6504 ext->chunks = newlen;
6505 ext->offset[id] = newoff;
6506 skb->extensions = ext;
6507 skb->active_extensions = 1 << id;
6508 return skb_ext_get_ptr(ext, id);
6512 * skb_ext_add - allocate space for given extension, COW if needed
6514 * @id: extension to allocate space for
6516 * Allocates enough space for the given extension.
6517 * If the extension is already present, a pointer to that extension
6520 * If the skb was cloned, COW applies and the returned memory can be
6521 * modified without changing the extension space of clones buffers.
6523 * Returns pointer to the extension or NULL on allocation failure.
6525 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6527 struct skb_ext *new, *old = NULL;
6528 unsigned int newlen, newoff;
6530 if (skb->active_extensions) {
6531 old = skb->extensions;
6533 new = skb_ext_maybe_cow(old, skb->active_extensions);
6537 if (__skb_ext_exist(new, id))
6540 newoff = new->chunks;
6542 newoff = SKB_EXT_CHUNKSIZEOF(*new);
6544 new = __skb_ext_alloc(GFP_ATOMIC);
6549 newlen = newoff + skb_ext_type_len[id];
6550 new->chunks = newlen;
6551 new->offset[id] = newoff;
6554 skb->extensions = new;
6555 skb->active_extensions |= 1 << id;
6556 return skb_ext_get_ptr(new, id);
6558 EXPORT_SYMBOL(skb_ext_add);
6561 static void skb_ext_put_sp(struct sec_path *sp)
6565 for (i = 0; i < sp->len; i++)
6566 xfrm_state_put(sp->xvec[i]);
6570 #ifdef CONFIG_MCTP_FLOWS
6571 static void skb_ext_put_mctp(struct mctp_flow *flow)
6574 mctp_key_unref(flow->key);
6578 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6580 struct skb_ext *ext = skb->extensions;
6582 skb->active_extensions &= ~(1 << id);
6583 if (skb->active_extensions == 0) {
6584 skb->extensions = NULL;
6587 } else if (id == SKB_EXT_SEC_PATH &&
6588 refcount_read(&ext->refcnt) == 1) {
6589 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6596 EXPORT_SYMBOL(__skb_ext_del);
6598 void __skb_ext_put(struct skb_ext *ext)
6600 /* If this is last clone, nothing can increment
6601 * it after check passes. Avoids one atomic op.
6603 if (refcount_read(&ext->refcnt) == 1)
6606 if (!refcount_dec_and_test(&ext->refcnt))
6610 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6611 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6613 #ifdef CONFIG_MCTP_FLOWS
6614 if (__skb_ext_exist(ext, SKB_EXT_MCTP))
6615 skb_ext_put_mctp(skb_ext_get_ptr(ext, SKB_EXT_MCTP));
6618 kmem_cache_free(skbuff_ext_cache, ext);
6620 EXPORT_SYMBOL(__skb_ext_put);
6621 #endif /* CONFIG_SKB_EXTENSIONS */
6624 * skb_attempt_defer_free - queue skb for remote freeing
6627 * Put @skb in a per-cpu list, using the cpu which
6628 * allocated the skb/pages to reduce false sharing
6629 * and memory zone spinlock contention.
6631 void skb_attempt_defer_free(struct sk_buff *skb)
6633 int cpu = skb->alloc_cpu;
6634 struct softnet_data *sd;
6635 unsigned long flags;
6636 unsigned int defer_max;
6639 if (WARN_ON_ONCE(cpu >= nr_cpu_ids) ||
6641 cpu == raw_smp_processor_id()) {
6642 nodefer: __kfree_skb(skb);
6646 sd = &per_cpu(softnet_data, cpu);
6647 defer_max = READ_ONCE(sysctl_skb_defer_max);
6648 if (READ_ONCE(sd->defer_count) >= defer_max)
6651 spin_lock_irqsave(&sd->defer_lock, flags);
6652 /* Send an IPI every time queue reaches half capacity. */
6653 kick = sd->defer_count == (defer_max >> 1);
6654 /* Paired with the READ_ONCE() few lines above */
6655 WRITE_ONCE(sd->defer_count, sd->defer_count + 1);
6657 skb->next = sd->defer_list;
6658 /* Paired with READ_ONCE() in skb_defer_free_flush() */
6659 WRITE_ONCE(sd->defer_list, skb);
6660 spin_unlock_irqrestore(&sd->defer_lock, flags);
6662 /* Make sure to trigger NET_RX_SOFTIRQ on the remote CPU
6663 * if we are unlucky enough (this seems very unlikely).
6665 if (unlikely(kick) && !cmpxchg(&sd->defer_ipi_scheduled, 0, 1))
6666 smp_call_function_single_async(cpu, &sd->defer_csd);